blob: d017283ae6f56fa8ea0ba1eeaa04077d443eb0fb [file] [log] [blame]
Chris Masone02119d2008-09-05 16:13:11 -04001/*
2 * Copyright (C) 2008 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19#include <linux/sched.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090020#include <linux/slab.h>
Chris Masone02119d2008-09-05 16:13:11 -040021#include "ctree.h"
22#include "transaction.h"
23#include "disk-io.h"
24#include "locking.h"
25#include "print-tree.h"
26#include "compat.h"
Christoph Hellwigb2950862008-12-02 09:54:17 -050027#include "tree-log.h"
Chris Masone02119d2008-09-05 16:13:11 -040028
29/* magic values for the inode_only field in btrfs_log_inode:
30 *
31 * LOG_INODE_ALL means to log everything
32 * LOG_INODE_EXISTS means to log just enough to recreate the inode
33 * during log replay
34 */
35#define LOG_INODE_ALL 0
36#define LOG_INODE_EXISTS 1
37
38/*
Chris Mason12fcfd22009-03-24 10:24:20 -040039 * directory trouble cases
40 *
41 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
42 * log, we must force a full commit before doing an fsync of the directory
43 * where the unlink was done.
44 * ---> record transid of last unlink/rename per directory
45 *
46 * mkdir foo/some_dir
47 * normal commit
48 * rename foo/some_dir foo2/some_dir
49 * mkdir foo/some_dir
50 * fsync foo/some_dir/some_file
51 *
52 * The fsync above will unlink the original some_dir without recording
53 * it in its new location (foo2). After a crash, some_dir will be gone
54 * unless the fsync of some_file forces a full commit
55 *
56 * 2) we must log any new names for any file or dir that is in the fsync
57 * log. ---> check inode while renaming/linking.
58 *
59 * 2a) we must log any new names for any file or dir during rename
60 * when the directory they are being removed from was logged.
61 * ---> check inode and old parent dir during rename
62 *
63 * 2a is actually the more important variant. With the extra logging
64 * a crash might unlink the old name without recreating the new one
65 *
66 * 3) after a crash, we must go through any directories with a link count
67 * of zero and redo the rm -rf
68 *
69 * mkdir f1/foo
70 * normal commit
71 * rm -rf f1/foo
72 * fsync(f1)
73 *
74 * The directory f1 was fully removed from the FS, but fsync was never
75 * called on f1, only its parent dir. After a crash the rm -rf must
76 * be replayed. This must be able to recurse down the entire
77 * directory tree. The inode link count fixup code takes care of the
78 * ugly details.
79 */
80
81/*
Chris Masone02119d2008-09-05 16:13:11 -040082 * stages for the tree walking. The first
83 * stage (0) is to only pin down the blocks we find
84 * the second stage (1) is to make sure that all the inodes
85 * we find in the log are created in the subvolume.
86 *
87 * The last stage is to deal with directories and links and extents
88 * and all the other fun semantics
89 */
90#define LOG_WALK_PIN_ONLY 0
91#define LOG_WALK_REPLAY_INODES 1
92#define LOG_WALK_REPLAY_ALL 2
93
Chris Mason12fcfd22009-03-24 10:24:20 -040094static int btrfs_log_inode(struct btrfs_trans_handle *trans,
Chris Masone02119d2008-09-05 16:13:11 -040095 struct btrfs_root *root, struct inode *inode,
96 int inode_only);
Yan Zhengec051c02009-01-05 15:43:42 -050097static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
98 struct btrfs_root *root,
99 struct btrfs_path *path, u64 objectid);
Chris Mason12fcfd22009-03-24 10:24:20 -0400100static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
101 struct btrfs_root *root,
102 struct btrfs_root *log,
103 struct btrfs_path *path,
104 u64 dirid, int del_all);
Chris Masone02119d2008-09-05 16:13:11 -0400105
106/*
107 * tree logging is a special write ahead log used to make sure that
108 * fsyncs and O_SYNCs can happen without doing full tree commits.
109 *
110 * Full tree commits are expensive because they require commonly
111 * modified blocks to be recowed, creating many dirty pages in the
112 * extent tree an 4x-6x higher write load than ext3.
113 *
114 * Instead of doing a tree commit on every fsync, we use the
115 * key ranges and transaction ids to find items for a given file or directory
116 * that have changed in this transaction. Those items are copied into
117 * a special tree (one per subvolume root), that tree is written to disk
118 * and then the fsync is considered complete.
119 *
120 * After a crash, items are copied out of the log-tree back into the
121 * subvolume tree. Any file data extents found are recorded in the extent
122 * allocation tree, and the log-tree freed.
123 *
124 * The log tree is read three times, once to pin down all the extents it is
125 * using in ram and once, once to create all the inodes logged in the tree
126 * and once to do all the other items.
127 */
128
129/*
Chris Masone02119d2008-09-05 16:13:11 -0400130 * start a sub transaction and setup the log tree
131 * this increments the log tree writer count to make the people
132 * syncing the tree wait for us to finish
133 */
134static int start_log_trans(struct btrfs_trans_handle *trans,
135 struct btrfs_root *root)
136{
137 int ret;
Yan, Zheng4a500fd2010-05-16 10:49:59 -0400138 int err = 0;
Yan Zheng7237f182009-01-21 12:54:03 -0500139
140 mutex_lock(&root->log_mutex);
141 if (root->log_root) {
Josef Bacikff782e02009-10-08 15:30:04 -0400142 if (!root->log_start_pid) {
143 root->log_start_pid = current->pid;
144 root->log_multiple_pids = false;
145 } else if (root->log_start_pid != current->pid) {
146 root->log_multiple_pids = true;
147 }
148
Yan Zheng7237f182009-01-21 12:54:03 -0500149 root->log_batch++;
150 atomic_inc(&root->log_writers);
151 mutex_unlock(&root->log_mutex);
152 return 0;
153 }
Josef Bacikff782e02009-10-08 15:30:04 -0400154 root->log_multiple_pids = false;
155 root->log_start_pid = current->pid;
Chris Masone02119d2008-09-05 16:13:11 -0400156 mutex_lock(&root->fs_info->tree_log_mutex);
157 if (!root->fs_info->log_root_tree) {
158 ret = btrfs_init_log_root_tree(trans, root->fs_info);
Yan, Zheng4a500fd2010-05-16 10:49:59 -0400159 if (ret)
160 err = ret;
Chris Masone02119d2008-09-05 16:13:11 -0400161 }
Yan, Zheng4a500fd2010-05-16 10:49:59 -0400162 if (err == 0 && !root->log_root) {
Chris Masone02119d2008-09-05 16:13:11 -0400163 ret = btrfs_add_log_tree(trans, root);
Yan, Zheng4a500fd2010-05-16 10:49:59 -0400164 if (ret)
165 err = ret;
Chris Masone02119d2008-09-05 16:13:11 -0400166 }
Chris Masone02119d2008-09-05 16:13:11 -0400167 mutex_unlock(&root->fs_info->tree_log_mutex);
Yan Zheng7237f182009-01-21 12:54:03 -0500168 root->log_batch++;
169 atomic_inc(&root->log_writers);
170 mutex_unlock(&root->log_mutex);
Yan, Zheng4a500fd2010-05-16 10:49:59 -0400171 return err;
Chris Masone02119d2008-09-05 16:13:11 -0400172}
173
174/*
175 * returns 0 if there was a log transaction running and we were able
176 * to join, or returns -ENOENT if there were not transactions
177 * in progress
178 */
179static int join_running_log_trans(struct btrfs_root *root)
180{
181 int ret = -ENOENT;
182
183 smp_mb();
184 if (!root->log_root)
185 return -ENOENT;
186
Yan Zheng7237f182009-01-21 12:54:03 -0500187 mutex_lock(&root->log_mutex);
Chris Masone02119d2008-09-05 16:13:11 -0400188 if (root->log_root) {
189 ret = 0;
Yan Zheng7237f182009-01-21 12:54:03 -0500190 atomic_inc(&root->log_writers);
Chris Masone02119d2008-09-05 16:13:11 -0400191 }
Yan Zheng7237f182009-01-21 12:54:03 -0500192 mutex_unlock(&root->log_mutex);
Chris Masone02119d2008-09-05 16:13:11 -0400193 return ret;
194}
195
196/*
Chris Mason12fcfd22009-03-24 10:24:20 -0400197 * This either makes the current running log transaction wait
198 * until you call btrfs_end_log_trans() or it makes any future
199 * log transactions wait until you call btrfs_end_log_trans()
200 */
201int btrfs_pin_log_trans(struct btrfs_root *root)
202{
203 int ret = -ENOENT;
204
205 mutex_lock(&root->log_mutex);
206 atomic_inc(&root->log_writers);
207 mutex_unlock(&root->log_mutex);
208 return ret;
209}
210
211/*
Chris Masone02119d2008-09-05 16:13:11 -0400212 * indicate we're done making changes to the log tree
213 * and wake up anyone waiting to do a sync
214 */
Jeff Mahoney143bede2012-03-01 14:56:26 +0100215void btrfs_end_log_trans(struct btrfs_root *root)
Chris Masone02119d2008-09-05 16:13:11 -0400216{
Yan Zheng7237f182009-01-21 12:54:03 -0500217 if (atomic_dec_and_test(&root->log_writers)) {
218 smp_mb();
219 if (waitqueue_active(&root->log_writer_wait))
220 wake_up(&root->log_writer_wait);
221 }
Chris Masone02119d2008-09-05 16:13:11 -0400222}
223
224
225/*
226 * the walk control struct is used to pass state down the chain when
227 * processing the log tree. The stage field tells us which part
228 * of the log tree processing we are currently doing. The others
229 * are state fields used for that specific part
230 */
231struct walk_control {
232 /* should we free the extent on disk when done? This is used
233 * at transaction commit time while freeing a log tree
234 */
235 int free;
236
237 /* should we write out the extent buffer? This is used
238 * while flushing the log tree to disk during a sync
239 */
240 int write;
241
242 /* should we wait for the extent buffer io to finish? Also used
243 * while flushing the log tree to disk for a sync
244 */
245 int wait;
246
247 /* pin only walk, we record which extents on disk belong to the
248 * log trees
249 */
250 int pin;
251
252 /* what stage of the replay code we're currently in */
253 int stage;
254
255 /* the root we are currently replaying */
256 struct btrfs_root *replay_dest;
257
258 /* the trans handle for the current replay */
259 struct btrfs_trans_handle *trans;
260
261 /* the function that gets used to process blocks we find in the
262 * tree. Note the extent_buffer might not be up to date when it is
263 * passed in, and it must be checked or read if you need the data
264 * inside it
265 */
266 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
267 struct walk_control *wc, u64 gen);
268};
269
270/*
271 * process_func used to pin down extents, write them or wait on them
272 */
273static int process_one_buffer(struct btrfs_root *log,
274 struct extent_buffer *eb,
275 struct walk_control *wc, u64 gen)
276{
Josef Bacik04018de2009-04-03 10:14:18 -0400277 if (wc->pin)
Chris Masone688b7252011-10-31 20:52:39 -0400278 btrfs_pin_extent_for_log_replay(wc->trans,
279 log->fs_info->extent_root,
280 eb->start, eb->len);
Chris Masone02119d2008-09-05 16:13:11 -0400281
282 if (btrfs_buffer_uptodate(eb, gen)) {
283 if (wc->write)
284 btrfs_write_tree_block(eb);
285 if (wc->wait)
286 btrfs_wait_tree_block_writeback(eb);
287 }
288 return 0;
289}
290
291/*
292 * Item overwrite used by replay and tree logging. eb, slot and key all refer
293 * to the src data we are copying out.
294 *
295 * root is the tree we are copying into, and path is a scratch
296 * path for use in this function (it should be released on entry and
297 * will be released on exit).
298 *
299 * If the key is already in the destination tree the existing item is
300 * overwritten. If the existing item isn't big enough, it is extended.
301 * If it is too large, it is truncated.
302 *
303 * If the key isn't in the destination yet, a new item is inserted.
304 */
305static noinline int overwrite_item(struct btrfs_trans_handle *trans,
306 struct btrfs_root *root,
307 struct btrfs_path *path,
308 struct extent_buffer *eb, int slot,
309 struct btrfs_key *key)
310{
311 int ret;
312 u32 item_size;
313 u64 saved_i_size = 0;
314 int save_old_i_size = 0;
315 unsigned long src_ptr;
316 unsigned long dst_ptr;
317 int overwrite_root = 0;
318
319 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
320 overwrite_root = 1;
321
322 item_size = btrfs_item_size_nr(eb, slot);
323 src_ptr = btrfs_item_ptr_offset(eb, slot);
324
325 /* look for the key in the destination tree */
326 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
327 if (ret == 0) {
328 char *src_copy;
329 char *dst_copy;
330 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
331 path->slots[0]);
332 if (dst_size != item_size)
333 goto insert;
334
335 if (item_size == 0) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200336 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400337 return 0;
338 }
339 dst_copy = kmalloc(item_size, GFP_NOFS);
340 src_copy = kmalloc(item_size, GFP_NOFS);
liubo2a29edc2011-01-26 06:22:08 +0000341 if (!dst_copy || !src_copy) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200342 btrfs_release_path(path);
liubo2a29edc2011-01-26 06:22:08 +0000343 kfree(dst_copy);
344 kfree(src_copy);
345 return -ENOMEM;
346 }
Chris Masone02119d2008-09-05 16:13:11 -0400347
348 read_extent_buffer(eb, src_copy, src_ptr, item_size);
349
350 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
351 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
352 item_size);
353 ret = memcmp(dst_copy, src_copy, item_size);
354
355 kfree(dst_copy);
356 kfree(src_copy);
357 /*
358 * they have the same contents, just return, this saves
359 * us from cowing blocks in the destination tree and doing
360 * extra writes that may not have been done by a previous
361 * sync
362 */
363 if (ret == 0) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200364 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400365 return 0;
366 }
367
368 }
369insert:
David Sterbab3b4aa72011-04-21 01:20:15 +0200370 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400371 /* try to insert the key into the destination tree */
372 ret = btrfs_insert_empty_item(trans, root, path,
373 key, item_size);
374
375 /* make sure any existing item is the correct size */
376 if (ret == -EEXIST) {
377 u32 found_size;
378 found_size = btrfs_item_size_nr(path->nodes[0],
379 path->slots[0]);
Jeff Mahoney143bede2012-03-01 14:56:26 +0100380 if (found_size > item_size)
Chris Masone02119d2008-09-05 16:13:11 -0400381 btrfs_truncate_item(trans, root, path, item_size, 1);
Jeff Mahoney143bede2012-03-01 14:56:26 +0100382 else if (found_size < item_size)
383 btrfs_extend_item(trans, root, path,
384 item_size - found_size);
Chris Masone02119d2008-09-05 16:13:11 -0400385 } else if (ret) {
Yan, Zheng4a500fd2010-05-16 10:49:59 -0400386 return ret;
Chris Masone02119d2008-09-05 16:13:11 -0400387 }
388 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
389 path->slots[0]);
390
391 /* don't overwrite an existing inode if the generation number
392 * was logged as zero. This is done when the tree logging code
393 * is just logging an inode to make sure it exists after recovery.
394 *
395 * Also, don't overwrite i_size on directories during replay.
396 * log replay inserts and removes directory items based on the
397 * state of the tree found in the subvolume, and i_size is modified
398 * as it goes
399 */
400 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
401 struct btrfs_inode_item *src_item;
402 struct btrfs_inode_item *dst_item;
403
404 src_item = (struct btrfs_inode_item *)src_ptr;
405 dst_item = (struct btrfs_inode_item *)dst_ptr;
406
407 if (btrfs_inode_generation(eb, src_item) == 0)
408 goto no_copy;
409
410 if (overwrite_root &&
411 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
412 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
413 save_old_i_size = 1;
414 saved_i_size = btrfs_inode_size(path->nodes[0],
415 dst_item);
416 }
417 }
418
419 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
420 src_ptr, item_size);
421
422 if (save_old_i_size) {
423 struct btrfs_inode_item *dst_item;
424 dst_item = (struct btrfs_inode_item *)dst_ptr;
425 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
426 }
427
428 /* make sure the generation is filled in */
429 if (key->type == BTRFS_INODE_ITEM_KEY) {
430 struct btrfs_inode_item *dst_item;
431 dst_item = (struct btrfs_inode_item *)dst_ptr;
432 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
433 btrfs_set_inode_generation(path->nodes[0], dst_item,
434 trans->transid);
435 }
436 }
437no_copy:
438 btrfs_mark_buffer_dirty(path->nodes[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +0200439 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400440 return 0;
441}
442
443/*
444 * simple helper to read an inode off the disk from a given root
445 * This can only be called for subvolume roots and not for the log
446 */
447static noinline struct inode *read_one_inode(struct btrfs_root *root,
448 u64 objectid)
449{
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400450 struct btrfs_key key;
Chris Masone02119d2008-09-05 16:13:11 -0400451 struct inode *inode;
Chris Masone02119d2008-09-05 16:13:11 -0400452
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400453 key.objectid = objectid;
454 key.type = BTRFS_INODE_ITEM_KEY;
455 key.offset = 0;
Josef Bacik73f73412009-12-04 17:38:27 +0000456 inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400457 if (IS_ERR(inode)) {
458 inode = NULL;
459 } else if (is_bad_inode(inode)) {
Chris Masone02119d2008-09-05 16:13:11 -0400460 iput(inode);
461 inode = NULL;
462 }
463 return inode;
464}
465
466/* replays a single extent in 'eb' at 'slot' with 'key' into the
467 * subvolume 'root'. path is released on entry and should be released
468 * on exit.
469 *
470 * extents in the log tree have not been allocated out of the extent
471 * tree yet. So, this completes the allocation, taking a reference
472 * as required if the extent already exists or creating a new extent
473 * if it isn't in the extent allocation tree yet.
474 *
475 * The extent is inserted into the file, dropping any existing extents
476 * from the file that overlap the new one.
477 */
478static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
479 struct btrfs_root *root,
480 struct btrfs_path *path,
481 struct extent_buffer *eb, int slot,
482 struct btrfs_key *key)
483{
484 int found_type;
485 u64 mask = root->sectorsize - 1;
486 u64 extent_end;
487 u64 alloc_hint;
488 u64 start = key->offset;
Yan Zheng07d400a2009-01-06 11:42:00 -0500489 u64 saved_nbytes;
Chris Masone02119d2008-09-05 16:13:11 -0400490 struct btrfs_file_extent_item *item;
491 struct inode *inode = NULL;
492 unsigned long size;
493 int ret = 0;
494
495 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
496 found_type = btrfs_file_extent_type(eb, item);
497
Yan Zhengd899e052008-10-30 14:25:28 -0400498 if (found_type == BTRFS_FILE_EXTENT_REG ||
499 found_type == BTRFS_FILE_EXTENT_PREALLOC)
Chris Masone02119d2008-09-05 16:13:11 -0400500 extent_end = start + btrfs_file_extent_num_bytes(eb, item);
501 else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
Chris Masonc8b97812008-10-29 14:49:59 -0400502 size = btrfs_file_extent_inline_len(eb, item);
Chris Masone02119d2008-09-05 16:13:11 -0400503 extent_end = (start + size + mask) & ~mask;
504 } else {
505 ret = 0;
506 goto out;
507 }
508
509 inode = read_one_inode(root, key->objectid);
510 if (!inode) {
511 ret = -EIO;
512 goto out;
513 }
514
515 /*
516 * first check to see if we already have this extent in the
517 * file. This must be done before the btrfs_drop_extents run
518 * so we don't try to drop this extent.
519 */
Li Zefan33345d012011-04-20 10:31:50 +0800520 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
Chris Masone02119d2008-09-05 16:13:11 -0400521 start, 0);
522
Yan Zhengd899e052008-10-30 14:25:28 -0400523 if (ret == 0 &&
524 (found_type == BTRFS_FILE_EXTENT_REG ||
525 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
Chris Masone02119d2008-09-05 16:13:11 -0400526 struct btrfs_file_extent_item cmp1;
527 struct btrfs_file_extent_item cmp2;
528 struct btrfs_file_extent_item *existing;
529 struct extent_buffer *leaf;
530
531 leaf = path->nodes[0];
532 existing = btrfs_item_ptr(leaf, path->slots[0],
533 struct btrfs_file_extent_item);
534
535 read_extent_buffer(eb, &cmp1, (unsigned long)item,
536 sizeof(cmp1));
537 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
538 sizeof(cmp2));
539
540 /*
541 * we already have a pointer to this exact extent,
542 * we don't have to do anything
543 */
544 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200545 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400546 goto out;
547 }
548 }
David Sterbab3b4aa72011-04-21 01:20:15 +0200549 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400550
Yan Zheng07d400a2009-01-06 11:42:00 -0500551 saved_nbytes = inode_get_bytes(inode);
Chris Masone02119d2008-09-05 16:13:11 -0400552 /* drop any overlapping extents */
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000553 ret = btrfs_drop_extents(trans, inode, start, extent_end,
554 &alloc_hint, 1);
Chris Masone02119d2008-09-05 16:13:11 -0400555 BUG_ON(ret);
556
Yan Zheng07d400a2009-01-06 11:42:00 -0500557 if (found_type == BTRFS_FILE_EXTENT_REG ||
558 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400559 u64 offset;
Yan Zheng07d400a2009-01-06 11:42:00 -0500560 unsigned long dest_offset;
561 struct btrfs_key ins;
Chris Masone02119d2008-09-05 16:13:11 -0400562
Yan Zheng07d400a2009-01-06 11:42:00 -0500563 ret = btrfs_insert_empty_item(trans, root, path, key,
564 sizeof(*item));
565 BUG_ON(ret);
566 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
567 path->slots[0]);
568 copy_extent_buffer(path->nodes[0], eb, dest_offset,
569 (unsigned long)item, sizeof(*item));
570
571 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
572 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
573 ins.type = BTRFS_EXTENT_ITEM_KEY;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400574 offset = key->offset - btrfs_file_extent_offset(eb, item);
Yan Zheng07d400a2009-01-06 11:42:00 -0500575
576 if (ins.objectid > 0) {
577 u64 csum_start;
578 u64 csum_end;
579 LIST_HEAD(ordered_sums);
580 /*
581 * is this extent already allocated in the extent
582 * allocation tree? If so, just add a reference
583 */
584 ret = btrfs_lookup_extent(root, ins.objectid,
585 ins.offset);
586 if (ret == 0) {
587 ret = btrfs_inc_extent_ref(trans, root,
588 ins.objectid, ins.offset,
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400589 0, root->root_key.objectid,
Arne Jansen66d7e7f2011-09-12 15:26:38 +0200590 key->objectid, offset, 0);
Tsutomu Itoh37daa4f2011-04-28 09:18:21 +0000591 BUG_ON(ret);
Yan Zheng07d400a2009-01-06 11:42:00 -0500592 } else {
593 /*
594 * insert the extent pointer in the extent
595 * allocation tree
596 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400597 ret = btrfs_alloc_logged_file_extent(trans,
598 root, root->root_key.objectid,
599 key->objectid, offset, &ins);
Yan Zheng07d400a2009-01-06 11:42:00 -0500600 BUG_ON(ret);
601 }
David Sterbab3b4aa72011-04-21 01:20:15 +0200602 btrfs_release_path(path);
Yan Zheng07d400a2009-01-06 11:42:00 -0500603
604 if (btrfs_file_extent_compression(eb, item)) {
605 csum_start = ins.objectid;
606 csum_end = csum_start + ins.offset;
607 } else {
608 csum_start = ins.objectid +
609 btrfs_file_extent_offset(eb, item);
610 csum_end = csum_start +
611 btrfs_file_extent_num_bytes(eb, item);
612 }
613
614 ret = btrfs_lookup_csums_range(root->log_root,
615 csum_start, csum_end - 1,
Arne Jansena2de7332011-03-08 14:14:00 +0100616 &ordered_sums, 0);
Yan Zheng07d400a2009-01-06 11:42:00 -0500617 BUG_ON(ret);
618 while (!list_empty(&ordered_sums)) {
619 struct btrfs_ordered_sum *sums;
620 sums = list_entry(ordered_sums.next,
621 struct btrfs_ordered_sum,
622 list);
623 ret = btrfs_csum_file_blocks(trans,
624 root->fs_info->csum_root,
625 sums);
626 BUG_ON(ret);
627 list_del(&sums->list);
628 kfree(sums);
629 }
630 } else {
David Sterbab3b4aa72011-04-21 01:20:15 +0200631 btrfs_release_path(path);
Yan Zheng07d400a2009-01-06 11:42:00 -0500632 }
633 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
634 /* inline extents are easy, we just overwrite them */
635 ret = overwrite_item(trans, root, path, eb, slot, key);
636 BUG_ON(ret);
637 }
638
639 inode_set_bytes(inode, saved_nbytes);
Chris Masone02119d2008-09-05 16:13:11 -0400640 btrfs_update_inode(trans, root, inode);
641out:
642 if (inode)
643 iput(inode);
644 return ret;
645}
646
647/*
648 * when cleaning up conflicts between the directory names in the
649 * subvolume, directory names in the log and directory names in the
650 * inode back references, we may have to unlink inodes from directories.
651 *
652 * This is a helper function to do the unlink of a specific directory
653 * item
654 */
655static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
656 struct btrfs_root *root,
657 struct btrfs_path *path,
658 struct inode *dir,
659 struct btrfs_dir_item *di)
660{
661 struct inode *inode;
662 char *name;
663 int name_len;
664 struct extent_buffer *leaf;
665 struct btrfs_key location;
666 int ret;
667
668 leaf = path->nodes[0];
669
670 btrfs_dir_item_key_to_cpu(leaf, di, &location);
671 name_len = btrfs_dir_name_len(leaf, di);
672 name = kmalloc(name_len, GFP_NOFS);
liubo2a29edc2011-01-26 06:22:08 +0000673 if (!name)
674 return -ENOMEM;
675
Chris Masone02119d2008-09-05 16:13:11 -0400676 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
David Sterbab3b4aa72011-04-21 01:20:15 +0200677 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400678
679 inode = read_one_inode(root, location.objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +0000680 if (!inode) {
681 kfree(name);
682 return -EIO;
683 }
Chris Masone02119d2008-09-05 16:13:11 -0400684
Yan Zhengec051c02009-01-05 15:43:42 -0500685 ret = link_to_fixup_dir(trans, root, path, location.objectid);
686 BUG_ON(ret);
Chris Mason12fcfd22009-03-24 10:24:20 -0400687
Chris Masone02119d2008-09-05 16:13:11 -0400688 ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
Yan Zhengec051c02009-01-05 15:43:42 -0500689 BUG_ON(ret);
Chris Masone02119d2008-09-05 16:13:11 -0400690 kfree(name);
691
692 iput(inode);
693 return ret;
694}
695
696/*
697 * helper function to see if a given name and sequence number found
698 * in an inode back reference are already in a directory and correctly
699 * point to this inode
700 */
701static noinline int inode_in_dir(struct btrfs_root *root,
702 struct btrfs_path *path,
703 u64 dirid, u64 objectid, u64 index,
704 const char *name, int name_len)
705{
706 struct btrfs_dir_item *di;
707 struct btrfs_key location;
708 int match = 0;
709
710 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
711 index, name, name_len, 0);
712 if (di && !IS_ERR(di)) {
713 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
714 if (location.objectid != objectid)
715 goto out;
716 } else
717 goto out;
David Sterbab3b4aa72011-04-21 01:20:15 +0200718 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400719
720 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
721 if (di && !IS_ERR(di)) {
722 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
723 if (location.objectid != objectid)
724 goto out;
725 } else
726 goto out;
727 match = 1;
728out:
David Sterbab3b4aa72011-04-21 01:20:15 +0200729 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400730 return match;
731}
732
733/*
734 * helper function to check a log tree for a named back reference in
735 * an inode. This is used to decide if a back reference that is
736 * found in the subvolume conflicts with what we find in the log.
737 *
738 * inode backreferences may have multiple refs in a single item,
739 * during replay we process one reference at a time, and we don't
740 * want to delete valid links to a file from the subvolume if that
741 * link is also in the log.
742 */
743static noinline int backref_in_log(struct btrfs_root *log,
744 struct btrfs_key *key,
745 char *name, int namelen)
746{
747 struct btrfs_path *path;
748 struct btrfs_inode_ref *ref;
749 unsigned long ptr;
750 unsigned long ptr_end;
751 unsigned long name_ptr;
752 int found_name_len;
753 int item_size;
754 int ret;
755 int match = 0;
756
757 path = btrfs_alloc_path();
liubo2a29edc2011-01-26 06:22:08 +0000758 if (!path)
759 return -ENOMEM;
760
Chris Masone02119d2008-09-05 16:13:11 -0400761 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
762 if (ret != 0)
763 goto out;
764
765 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
766 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
767 ptr_end = ptr + item_size;
768 while (ptr < ptr_end) {
769 ref = (struct btrfs_inode_ref *)ptr;
770 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
771 if (found_name_len == namelen) {
772 name_ptr = (unsigned long)(ref + 1);
773 ret = memcmp_extent_buffer(path->nodes[0], name,
774 name_ptr, namelen);
775 if (ret == 0) {
776 match = 1;
777 goto out;
778 }
779 }
780 ptr = (unsigned long)(ref + 1) + found_name_len;
781 }
782out:
783 btrfs_free_path(path);
784 return match;
785}
786
787
788/*
789 * replay one inode back reference item found in the log tree.
790 * eb, slot and key refer to the buffer and key found in the log tree.
791 * root is the destination we are replaying into, and path is for temp
792 * use by this function. (it should be released on return).
793 */
794static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
795 struct btrfs_root *root,
796 struct btrfs_root *log,
797 struct btrfs_path *path,
798 struct extent_buffer *eb, int slot,
799 struct btrfs_key *key)
800{
Chris Masone02119d2008-09-05 16:13:11 -0400801 struct btrfs_inode_ref *ref;
liubo34f3e4f2011-08-06 08:35:23 +0000802 struct btrfs_dir_item *di;
803 struct inode *dir;
Chris Masone02119d2008-09-05 16:13:11 -0400804 struct inode *inode;
Chris Masone02119d2008-09-05 16:13:11 -0400805 unsigned long ref_ptr;
806 unsigned long ref_end;
liubo34f3e4f2011-08-06 08:35:23 +0000807 char *name;
808 int namelen;
809 int ret;
liuboc622ae62011-03-26 08:01:12 -0400810 int search_done = 0;
Chris Masone02119d2008-09-05 16:13:11 -0400811
Chris Masone02119d2008-09-05 16:13:11 -0400812 /*
813 * it is possible that we didn't log all the parent directories
814 * for a given inode. If we don't find the dir, just don't
815 * copy the back ref in. The link count fixup code will take
816 * care of the rest
817 */
818 dir = read_one_inode(root, key->offset);
819 if (!dir)
820 return -ENOENT;
821
822 inode = read_one_inode(root, key->objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +0000823 if (!inode) {
824 iput(dir);
825 return -EIO;
826 }
Chris Masone02119d2008-09-05 16:13:11 -0400827
828 ref_ptr = btrfs_item_ptr_offset(eb, slot);
829 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
830
831again:
832 ref = (struct btrfs_inode_ref *)ref_ptr;
833
834 namelen = btrfs_inode_ref_name_len(eb, ref);
835 name = kmalloc(namelen, GFP_NOFS);
836 BUG_ON(!name);
837
838 read_extent_buffer(eb, name, (unsigned long)(ref + 1), namelen);
839
840 /* if we already have a perfect match, we're done */
Li Zefan33345d012011-04-20 10:31:50 +0800841 if (inode_in_dir(root, path, btrfs_ino(dir), btrfs_ino(inode),
Chris Masone02119d2008-09-05 16:13:11 -0400842 btrfs_inode_ref_index(eb, ref),
843 name, namelen)) {
844 goto out;
845 }
846
847 /*
848 * look for a conflicting back reference in the metadata.
849 * if we find one we have to unlink that name of the file
850 * before we add our new link. Later on, we overwrite any
851 * existing back reference, and we don't want to create
852 * dangling pointers in the directory.
853 */
liuboc622ae62011-03-26 08:01:12 -0400854
855 if (search_done)
856 goto insert;
857
Chris Masone02119d2008-09-05 16:13:11 -0400858 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
859 if (ret == 0) {
860 char *victim_name;
861 int victim_name_len;
862 struct btrfs_inode_ref *victim_ref;
863 unsigned long ptr;
864 unsigned long ptr_end;
865 struct extent_buffer *leaf = path->nodes[0];
866
867 /* are we trying to overwrite a back ref for the root directory
868 * if so, just jump out, we're done
869 */
870 if (key->objectid == key->offset)
871 goto out_nowrite;
872
873 /* check all the names in this back reference to see
874 * if they are in the log. if so, we allow them to stay
875 * otherwise they must be unlinked as a conflict
876 */
877 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
878 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
Chris Masond3977122009-01-05 21:25:51 -0500879 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -0400880 victim_ref = (struct btrfs_inode_ref *)ptr;
881 victim_name_len = btrfs_inode_ref_name_len(leaf,
882 victim_ref);
883 victim_name = kmalloc(victim_name_len, GFP_NOFS);
884 BUG_ON(!victim_name);
885
886 read_extent_buffer(leaf, victim_name,
887 (unsigned long)(victim_ref + 1),
888 victim_name_len);
889
890 if (!backref_in_log(log, key, victim_name,
891 victim_name_len)) {
892 btrfs_inc_nlink(inode);
David Sterbab3b4aa72011-04-21 01:20:15 +0200893 btrfs_release_path(path);
Chris Mason12fcfd22009-03-24 10:24:20 -0400894
Chris Masone02119d2008-09-05 16:13:11 -0400895 ret = btrfs_unlink_inode(trans, root, dir,
896 inode, victim_name,
897 victim_name_len);
Chris Masone02119d2008-09-05 16:13:11 -0400898 }
899 kfree(victim_name);
900 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
901 }
902 BUG_ON(ret);
liuboc622ae62011-03-26 08:01:12 -0400903
904 /*
905 * NOTE: we have searched root tree and checked the
906 * coresponding ref, it does not need to check again.
907 */
908 search_done = 1;
Chris Masone02119d2008-09-05 16:13:11 -0400909 }
David Sterbab3b4aa72011-04-21 01:20:15 +0200910 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400911
liubo34f3e4f2011-08-06 08:35:23 +0000912 /* look for a conflicting sequence number */
913 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
914 btrfs_inode_ref_index(eb, ref),
915 name, namelen, 0);
916 if (di && !IS_ERR(di)) {
917 ret = drop_one_dir_item(trans, root, path, dir, di);
918 BUG_ON(ret);
919 }
920 btrfs_release_path(path);
921
922 /* look for a conflicing name */
923 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
924 name, namelen, 0);
925 if (di && !IS_ERR(di)) {
926 ret = drop_one_dir_item(trans, root, path, dir, di);
927 BUG_ON(ret);
928 }
929 btrfs_release_path(path);
930
liuboc622ae62011-03-26 08:01:12 -0400931insert:
Chris Masone02119d2008-09-05 16:13:11 -0400932 /* insert our name */
933 ret = btrfs_add_link(trans, dir, inode, name, namelen, 0,
934 btrfs_inode_ref_index(eb, ref));
935 BUG_ON(ret);
936
937 btrfs_update_inode(trans, root, inode);
938
939out:
940 ref_ptr = (unsigned long)(ref + 1) + namelen;
941 kfree(name);
942 if (ref_ptr < ref_end)
943 goto again;
944
945 /* finally write the back reference in the inode */
946 ret = overwrite_item(trans, root, path, eb, slot, key);
947 BUG_ON(ret);
948
949out_nowrite:
David Sterbab3b4aa72011-04-21 01:20:15 +0200950 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -0400951 iput(dir);
952 iput(inode);
953 return 0;
954}
955
Yan, Zhengc71bf092009-11-12 09:34:40 +0000956static int insert_orphan_item(struct btrfs_trans_handle *trans,
957 struct btrfs_root *root, u64 offset)
958{
959 int ret;
960 ret = btrfs_find_orphan_item(root, offset);
961 if (ret > 0)
962 ret = btrfs_insert_orphan_item(trans, root, offset);
963 return ret;
964}
965
966
Chris Masone02119d2008-09-05 16:13:11 -0400967/*
Chris Masone02119d2008-09-05 16:13:11 -0400968 * There are a few corners where the link count of the file can't
969 * be properly maintained during replay. So, instead of adding
970 * lots of complexity to the log code, we just scan the backrefs
971 * for any file that has been through replay.
972 *
973 * The scan will update the link count on the inode to reflect the
974 * number of back refs found. If it goes down to zero, the iput
975 * will free the inode.
976 */
977static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
978 struct btrfs_root *root,
979 struct inode *inode)
980{
981 struct btrfs_path *path;
982 int ret;
983 struct btrfs_key key;
984 u64 nlink = 0;
985 unsigned long ptr;
986 unsigned long ptr_end;
987 int name_len;
Li Zefan33345d012011-04-20 10:31:50 +0800988 u64 ino = btrfs_ino(inode);
Chris Masone02119d2008-09-05 16:13:11 -0400989
Li Zefan33345d012011-04-20 10:31:50 +0800990 key.objectid = ino;
Chris Masone02119d2008-09-05 16:13:11 -0400991 key.type = BTRFS_INODE_REF_KEY;
992 key.offset = (u64)-1;
993
994 path = btrfs_alloc_path();
liubo2a29edc2011-01-26 06:22:08 +0000995 if (!path)
996 return -ENOMEM;
Chris Masone02119d2008-09-05 16:13:11 -0400997
Chris Masond3977122009-01-05 21:25:51 -0500998 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -0400999 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1000 if (ret < 0)
1001 break;
1002 if (ret > 0) {
1003 if (path->slots[0] == 0)
1004 break;
1005 path->slots[0]--;
1006 }
1007 btrfs_item_key_to_cpu(path->nodes[0], &key,
1008 path->slots[0]);
Li Zefan33345d012011-04-20 10:31:50 +08001009 if (key.objectid != ino ||
Chris Masone02119d2008-09-05 16:13:11 -04001010 key.type != BTRFS_INODE_REF_KEY)
1011 break;
1012 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1013 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1014 path->slots[0]);
Chris Masond3977122009-01-05 21:25:51 -05001015 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -04001016 struct btrfs_inode_ref *ref;
1017
1018 ref = (struct btrfs_inode_ref *)ptr;
1019 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1020 ref);
1021 ptr = (unsigned long)(ref + 1) + name_len;
1022 nlink++;
1023 }
1024
1025 if (key.offset == 0)
1026 break;
1027 key.offset--;
David Sterbab3b4aa72011-04-21 01:20:15 +02001028 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001029 }
David Sterbab3b4aa72011-04-21 01:20:15 +02001030 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001031 if (nlink != inode->i_nlink) {
Miklos Szeredibfe86842011-10-28 14:13:29 +02001032 set_nlink(inode, nlink);
Chris Masone02119d2008-09-05 16:13:11 -04001033 btrfs_update_inode(trans, root, inode);
1034 }
Chris Mason8d5bf1c2008-09-11 15:51:21 -04001035 BTRFS_I(inode)->index_cnt = (u64)-1;
Chris Masone02119d2008-09-05 16:13:11 -04001036
Yan, Zhengc71bf092009-11-12 09:34:40 +00001037 if (inode->i_nlink == 0) {
1038 if (S_ISDIR(inode->i_mode)) {
1039 ret = replay_dir_deletes(trans, root, NULL, path,
Li Zefan33345d012011-04-20 10:31:50 +08001040 ino, 1);
Yan, Zhengc71bf092009-11-12 09:34:40 +00001041 BUG_ON(ret);
1042 }
Li Zefan33345d012011-04-20 10:31:50 +08001043 ret = insert_orphan_item(trans, root, ino);
Chris Mason12fcfd22009-03-24 10:24:20 -04001044 BUG_ON(ret);
1045 }
1046 btrfs_free_path(path);
1047
Chris Masone02119d2008-09-05 16:13:11 -04001048 return 0;
1049}
1050
1051static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1052 struct btrfs_root *root,
1053 struct btrfs_path *path)
1054{
1055 int ret;
1056 struct btrfs_key key;
1057 struct inode *inode;
1058
1059 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1060 key.type = BTRFS_ORPHAN_ITEM_KEY;
1061 key.offset = (u64)-1;
Chris Masond3977122009-01-05 21:25:51 -05001062 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04001063 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1064 if (ret < 0)
1065 break;
1066
1067 if (ret == 1) {
1068 if (path->slots[0] == 0)
1069 break;
1070 path->slots[0]--;
1071 }
1072
1073 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1074 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1075 key.type != BTRFS_ORPHAN_ITEM_KEY)
1076 break;
1077
1078 ret = btrfs_del_item(trans, root, path);
Tsutomu Itoh65a246c2011-05-19 04:37:44 +00001079 if (ret)
1080 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001081
David Sterbab3b4aa72011-04-21 01:20:15 +02001082 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001083 inode = read_one_inode(root, key.offset);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001084 if (!inode)
1085 return -EIO;
Chris Masone02119d2008-09-05 16:13:11 -04001086
1087 ret = fixup_inode_link_count(trans, root, inode);
1088 BUG_ON(ret);
1089
1090 iput(inode);
1091
Chris Mason12fcfd22009-03-24 10:24:20 -04001092 /*
1093 * fixup on a directory may create new entries,
1094 * make sure we always look for the highset possible
1095 * offset
1096 */
1097 key.offset = (u64)-1;
Chris Masone02119d2008-09-05 16:13:11 -04001098 }
Tsutomu Itoh65a246c2011-05-19 04:37:44 +00001099 ret = 0;
1100out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001101 btrfs_release_path(path);
Tsutomu Itoh65a246c2011-05-19 04:37:44 +00001102 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04001103}
1104
1105
1106/*
1107 * record a given inode in the fixup dir so we can check its link
1108 * count when replay is done. The link count is incremented here
1109 * so the inode won't go away until we check it
1110 */
1111static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1112 struct btrfs_root *root,
1113 struct btrfs_path *path,
1114 u64 objectid)
1115{
1116 struct btrfs_key key;
1117 int ret = 0;
1118 struct inode *inode;
1119
1120 inode = read_one_inode(root, objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001121 if (!inode)
1122 return -EIO;
Chris Masone02119d2008-09-05 16:13:11 -04001123
1124 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1125 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1126 key.offset = objectid;
1127
1128 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1129
David Sterbab3b4aa72011-04-21 01:20:15 +02001130 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001131 if (ret == 0) {
1132 btrfs_inc_nlink(inode);
1133 btrfs_update_inode(trans, root, inode);
1134 } else if (ret == -EEXIST) {
1135 ret = 0;
1136 } else {
1137 BUG();
1138 }
1139 iput(inode);
1140
1141 return ret;
1142}
1143
1144/*
1145 * when replaying the log for a directory, we only insert names
1146 * for inodes that actually exist. This means an fsync on a directory
1147 * does not implicitly fsync all the new files in it
1148 */
1149static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1150 struct btrfs_root *root,
1151 struct btrfs_path *path,
1152 u64 dirid, u64 index,
1153 char *name, int name_len, u8 type,
1154 struct btrfs_key *location)
1155{
1156 struct inode *inode;
1157 struct inode *dir;
1158 int ret;
1159
1160 inode = read_one_inode(root, location->objectid);
1161 if (!inode)
1162 return -ENOENT;
1163
1164 dir = read_one_inode(root, dirid);
1165 if (!dir) {
1166 iput(inode);
1167 return -EIO;
1168 }
1169 ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1170
1171 /* FIXME, put inode into FIXUP list */
1172
1173 iput(inode);
1174 iput(dir);
1175 return ret;
1176}
1177
1178/*
1179 * take a single entry in a log directory item and replay it into
1180 * the subvolume.
1181 *
1182 * if a conflicting item exists in the subdirectory already,
1183 * the inode it points to is unlinked and put into the link count
1184 * fix up tree.
1185 *
1186 * If a name from the log points to a file or directory that does
1187 * not exist in the FS, it is skipped. fsyncs on directories
1188 * do not force down inodes inside that directory, just changes to the
1189 * names or unlinks in a directory.
1190 */
1191static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1192 struct btrfs_root *root,
1193 struct btrfs_path *path,
1194 struct extent_buffer *eb,
1195 struct btrfs_dir_item *di,
1196 struct btrfs_key *key)
1197{
1198 char *name;
1199 int name_len;
1200 struct btrfs_dir_item *dst_di;
1201 struct btrfs_key found_key;
1202 struct btrfs_key log_key;
1203 struct inode *dir;
Chris Masone02119d2008-09-05 16:13:11 -04001204 u8 log_type;
Chris Mason4bef0842008-09-08 11:18:08 -04001205 int exists;
Chris Masone02119d2008-09-05 16:13:11 -04001206 int ret;
1207
1208 dir = read_one_inode(root, key->objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001209 if (!dir)
1210 return -EIO;
Chris Masone02119d2008-09-05 16:13:11 -04001211
1212 name_len = btrfs_dir_name_len(eb, di);
1213 name = kmalloc(name_len, GFP_NOFS);
liubo2a29edc2011-01-26 06:22:08 +00001214 if (!name)
1215 return -ENOMEM;
1216
Chris Masone02119d2008-09-05 16:13:11 -04001217 log_type = btrfs_dir_type(eb, di);
1218 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1219 name_len);
1220
1221 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
Chris Mason4bef0842008-09-08 11:18:08 -04001222 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1223 if (exists == 0)
1224 exists = 1;
1225 else
1226 exists = 0;
David Sterbab3b4aa72011-04-21 01:20:15 +02001227 btrfs_release_path(path);
Chris Mason4bef0842008-09-08 11:18:08 -04001228
Chris Masone02119d2008-09-05 16:13:11 -04001229 if (key->type == BTRFS_DIR_ITEM_KEY) {
1230 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1231 name, name_len, 1);
Chris Masond3977122009-01-05 21:25:51 -05001232 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04001233 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1234 key->objectid,
1235 key->offset, name,
1236 name_len, 1);
1237 } else {
1238 BUG();
1239 }
David Sterbac7040052011-04-19 18:00:01 +02001240 if (IS_ERR_OR_NULL(dst_di)) {
Chris Masone02119d2008-09-05 16:13:11 -04001241 /* we need a sequence number to insert, so we only
1242 * do inserts for the BTRFS_DIR_INDEX_KEY types
1243 */
1244 if (key->type != BTRFS_DIR_INDEX_KEY)
1245 goto out;
1246 goto insert;
1247 }
1248
1249 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1250 /* the existing item matches the logged item */
1251 if (found_key.objectid == log_key.objectid &&
1252 found_key.type == log_key.type &&
1253 found_key.offset == log_key.offset &&
1254 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1255 goto out;
1256 }
1257
1258 /*
1259 * don't drop the conflicting directory entry if the inode
1260 * for the new entry doesn't exist
1261 */
Chris Mason4bef0842008-09-08 11:18:08 -04001262 if (!exists)
Chris Masone02119d2008-09-05 16:13:11 -04001263 goto out;
1264
Chris Masone02119d2008-09-05 16:13:11 -04001265 ret = drop_one_dir_item(trans, root, path, dir, dst_di);
1266 BUG_ON(ret);
1267
1268 if (key->type == BTRFS_DIR_INDEX_KEY)
1269 goto insert;
1270out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001271 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001272 kfree(name);
1273 iput(dir);
1274 return 0;
1275
1276insert:
David Sterbab3b4aa72011-04-21 01:20:15 +02001277 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001278 ret = insert_one_name(trans, root, path, key->objectid, key->offset,
1279 name, name_len, log_type, &log_key);
1280
Stoyan Gaydarovc2934982009-04-02 17:05:11 -04001281 BUG_ON(ret && ret != -ENOENT);
Chris Masone02119d2008-09-05 16:13:11 -04001282 goto out;
1283}
1284
1285/*
1286 * find all the names in a directory item and reconcile them into
1287 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1288 * one name in a directory item, but the same code gets used for
1289 * both directory index types
1290 */
1291static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1292 struct btrfs_root *root,
1293 struct btrfs_path *path,
1294 struct extent_buffer *eb, int slot,
1295 struct btrfs_key *key)
1296{
1297 int ret;
1298 u32 item_size = btrfs_item_size_nr(eb, slot);
1299 struct btrfs_dir_item *di;
1300 int name_len;
1301 unsigned long ptr;
1302 unsigned long ptr_end;
1303
1304 ptr = btrfs_item_ptr_offset(eb, slot);
1305 ptr_end = ptr + item_size;
Chris Masond3977122009-01-05 21:25:51 -05001306 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -04001307 di = (struct btrfs_dir_item *)ptr;
Josef Bacik22a94d42011-03-16 16:47:17 -04001308 if (verify_dir_item(root, eb, di))
1309 return -EIO;
Chris Masone02119d2008-09-05 16:13:11 -04001310 name_len = btrfs_dir_name_len(eb, di);
1311 ret = replay_one_name(trans, root, path, eb, di, key);
1312 BUG_ON(ret);
1313 ptr = (unsigned long)(di + 1);
1314 ptr += name_len;
1315 }
1316 return 0;
1317}
1318
1319/*
1320 * directory replay has two parts. There are the standard directory
1321 * items in the log copied from the subvolume, and range items
1322 * created in the log while the subvolume was logged.
1323 *
1324 * The range items tell us which parts of the key space the log
1325 * is authoritative for. During replay, if a key in the subvolume
1326 * directory is in a logged range item, but not actually in the log
1327 * that means it was deleted from the directory before the fsync
1328 * and should be removed.
1329 */
1330static noinline int find_dir_range(struct btrfs_root *root,
1331 struct btrfs_path *path,
1332 u64 dirid, int key_type,
1333 u64 *start_ret, u64 *end_ret)
1334{
1335 struct btrfs_key key;
1336 u64 found_end;
1337 struct btrfs_dir_log_item *item;
1338 int ret;
1339 int nritems;
1340
1341 if (*start_ret == (u64)-1)
1342 return 1;
1343
1344 key.objectid = dirid;
1345 key.type = key_type;
1346 key.offset = *start_ret;
1347
1348 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1349 if (ret < 0)
1350 goto out;
1351 if (ret > 0) {
1352 if (path->slots[0] == 0)
1353 goto out;
1354 path->slots[0]--;
1355 }
1356 if (ret != 0)
1357 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1358
1359 if (key.type != key_type || key.objectid != dirid) {
1360 ret = 1;
1361 goto next;
1362 }
1363 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1364 struct btrfs_dir_log_item);
1365 found_end = btrfs_dir_log_end(path->nodes[0], item);
1366
1367 if (*start_ret >= key.offset && *start_ret <= found_end) {
1368 ret = 0;
1369 *start_ret = key.offset;
1370 *end_ret = found_end;
1371 goto out;
1372 }
1373 ret = 1;
1374next:
1375 /* check the next slot in the tree to see if it is a valid item */
1376 nritems = btrfs_header_nritems(path->nodes[0]);
1377 if (path->slots[0] >= nritems) {
1378 ret = btrfs_next_leaf(root, path);
1379 if (ret)
1380 goto out;
1381 } else {
1382 path->slots[0]++;
1383 }
1384
1385 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1386
1387 if (key.type != key_type || key.objectid != dirid) {
1388 ret = 1;
1389 goto out;
1390 }
1391 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1392 struct btrfs_dir_log_item);
1393 found_end = btrfs_dir_log_end(path->nodes[0], item);
1394 *start_ret = key.offset;
1395 *end_ret = found_end;
1396 ret = 0;
1397out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001398 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001399 return ret;
1400}
1401
1402/*
1403 * this looks for a given directory item in the log. If the directory
1404 * item is not in the log, the item is removed and the inode it points
1405 * to is unlinked
1406 */
1407static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1408 struct btrfs_root *root,
1409 struct btrfs_root *log,
1410 struct btrfs_path *path,
1411 struct btrfs_path *log_path,
1412 struct inode *dir,
1413 struct btrfs_key *dir_key)
1414{
1415 int ret;
1416 struct extent_buffer *eb;
1417 int slot;
1418 u32 item_size;
1419 struct btrfs_dir_item *di;
1420 struct btrfs_dir_item *log_di;
1421 int name_len;
1422 unsigned long ptr;
1423 unsigned long ptr_end;
1424 char *name;
1425 struct inode *inode;
1426 struct btrfs_key location;
1427
1428again:
1429 eb = path->nodes[0];
1430 slot = path->slots[0];
1431 item_size = btrfs_item_size_nr(eb, slot);
1432 ptr = btrfs_item_ptr_offset(eb, slot);
1433 ptr_end = ptr + item_size;
Chris Masond3977122009-01-05 21:25:51 -05001434 while (ptr < ptr_end) {
Chris Masone02119d2008-09-05 16:13:11 -04001435 di = (struct btrfs_dir_item *)ptr;
Josef Bacik22a94d42011-03-16 16:47:17 -04001436 if (verify_dir_item(root, eb, di)) {
1437 ret = -EIO;
1438 goto out;
1439 }
1440
Chris Masone02119d2008-09-05 16:13:11 -04001441 name_len = btrfs_dir_name_len(eb, di);
1442 name = kmalloc(name_len, GFP_NOFS);
1443 if (!name) {
1444 ret = -ENOMEM;
1445 goto out;
1446 }
1447 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1448 name_len);
1449 log_di = NULL;
Chris Mason12fcfd22009-03-24 10:24:20 -04001450 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04001451 log_di = btrfs_lookup_dir_item(trans, log, log_path,
1452 dir_key->objectid,
1453 name, name_len, 0);
Chris Mason12fcfd22009-03-24 10:24:20 -04001454 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
Chris Masone02119d2008-09-05 16:13:11 -04001455 log_di = btrfs_lookup_dir_index_item(trans, log,
1456 log_path,
1457 dir_key->objectid,
1458 dir_key->offset,
1459 name, name_len, 0);
1460 }
David Sterbac7040052011-04-19 18:00:01 +02001461 if (IS_ERR_OR_NULL(log_di)) {
Chris Masone02119d2008-09-05 16:13:11 -04001462 btrfs_dir_item_key_to_cpu(eb, di, &location);
David Sterbab3b4aa72011-04-21 01:20:15 +02001463 btrfs_release_path(path);
1464 btrfs_release_path(log_path);
Chris Masone02119d2008-09-05 16:13:11 -04001465 inode = read_one_inode(root, location.objectid);
Tsutomu Itohc00e9492011-04-28 09:10:23 +00001466 if (!inode) {
1467 kfree(name);
1468 return -EIO;
1469 }
Chris Masone02119d2008-09-05 16:13:11 -04001470
1471 ret = link_to_fixup_dir(trans, root,
1472 path, location.objectid);
1473 BUG_ON(ret);
1474 btrfs_inc_nlink(inode);
1475 ret = btrfs_unlink_inode(trans, root, dir, inode,
1476 name, name_len);
1477 BUG_ON(ret);
1478 kfree(name);
1479 iput(inode);
1480
1481 /* there might still be more names under this key
1482 * check and repeat if required
1483 */
1484 ret = btrfs_search_slot(NULL, root, dir_key, path,
1485 0, 0);
1486 if (ret == 0)
1487 goto again;
1488 ret = 0;
1489 goto out;
1490 }
David Sterbab3b4aa72011-04-21 01:20:15 +02001491 btrfs_release_path(log_path);
Chris Masone02119d2008-09-05 16:13:11 -04001492 kfree(name);
1493
1494 ptr = (unsigned long)(di + 1);
1495 ptr += name_len;
1496 }
1497 ret = 0;
1498out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001499 btrfs_release_path(path);
1500 btrfs_release_path(log_path);
Chris Masone02119d2008-09-05 16:13:11 -04001501 return ret;
1502}
1503
1504/*
1505 * deletion replay happens before we copy any new directory items
1506 * out of the log or out of backreferences from inodes. It
1507 * scans the log to find ranges of keys that log is authoritative for,
1508 * and then scans the directory to find items in those ranges that are
1509 * not present in the log.
1510 *
1511 * Anything we don't find in the log is unlinked and removed from the
1512 * directory.
1513 */
1514static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
1515 struct btrfs_root *root,
1516 struct btrfs_root *log,
1517 struct btrfs_path *path,
Chris Mason12fcfd22009-03-24 10:24:20 -04001518 u64 dirid, int del_all)
Chris Masone02119d2008-09-05 16:13:11 -04001519{
1520 u64 range_start;
1521 u64 range_end;
1522 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
1523 int ret = 0;
1524 struct btrfs_key dir_key;
1525 struct btrfs_key found_key;
1526 struct btrfs_path *log_path;
1527 struct inode *dir;
1528
1529 dir_key.objectid = dirid;
1530 dir_key.type = BTRFS_DIR_ITEM_KEY;
1531 log_path = btrfs_alloc_path();
1532 if (!log_path)
1533 return -ENOMEM;
1534
1535 dir = read_one_inode(root, dirid);
1536 /* it isn't an error if the inode isn't there, that can happen
1537 * because we replay the deletes before we copy in the inode item
1538 * from the log
1539 */
1540 if (!dir) {
1541 btrfs_free_path(log_path);
1542 return 0;
1543 }
1544again:
1545 range_start = 0;
1546 range_end = 0;
Chris Masond3977122009-01-05 21:25:51 -05001547 while (1) {
Chris Mason12fcfd22009-03-24 10:24:20 -04001548 if (del_all)
1549 range_end = (u64)-1;
1550 else {
1551 ret = find_dir_range(log, path, dirid, key_type,
1552 &range_start, &range_end);
1553 if (ret != 0)
1554 break;
1555 }
Chris Masone02119d2008-09-05 16:13:11 -04001556
1557 dir_key.offset = range_start;
Chris Masond3977122009-01-05 21:25:51 -05001558 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04001559 int nritems;
1560 ret = btrfs_search_slot(NULL, root, &dir_key, path,
1561 0, 0);
1562 if (ret < 0)
1563 goto out;
1564
1565 nritems = btrfs_header_nritems(path->nodes[0]);
1566 if (path->slots[0] >= nritems) {
1567 ret = btrfs_next_leaf(root, path);
1568 if (ret)
1569 break;
1570 }
1571 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1572 path->slots[0]);
1573 if (found_key.objectid != dirid ||
1574 found_key.type != dir_key.type)
1575 goto next_type;
1576
1577 if (found_key.offset > range_end)
1578 break;
1579
1580 ret = check_item_in_log(trans, root, log, path,
Chris Mason12fcfd22009-03-24 10:24:20 -04001581 log_path, dir,
1582 &found_key);
Chris Masone02119d2008-09-05 16:13:11 -04001583 BUG_ON(ret);
1584 if (found_key.offset == (u64)-1)
1585 break;
1586 dir_key.offset = found_key.offset + 1;
1587 }
David Sterbab3b4aa72011-04-21 01:20:15 +02001588 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001589 if (range_end == (u64)-1)
1590 break;
1591 range_start = range_end + 1;
1592 }
1593
1594next_type:
1595 ret = 0;
1596 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
1597 key_type = BTRFS_DIR_LOG_INDEX_KEY;
1598 dir_key.type = BTRFS_DIR_INDEX_KEY;
David Sterbab3b4aa72011-04-21 01:20:15 +02001599 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001600 goto again;
1601 }
1602out:
David Sterbab3b4aa72011-04-21 01:20:15 +02001603 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001604 btrfs_free_path(log_path);
1605 iput(dir);
1606 return ret;
1607}
1608
1609/*
1610 * the process_func used to replay items from the log tree. This
1611 * gets called in two different stages. The first stage just looks
1612 * for inodes and makes sure they are all copied into the subvolume.
1613 *
1614 * The second stage copies all the other item types from the log into
1615 * the subvolume. The two stage approach is slower, but gets rid of
1616 * lots of complexity around inodes referencing other inodes that exist
1617 * only in the log (references come from either directory items or inode
1618 * back refs).
1619 */
1620static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
1621 struct walk_control *wc, u64 gen)
1622{
1623 int nritems;
1624 struct btrfs_path *path;
1625 struct btrfs_root *root = wc->replay_dest;
1626 struct btrfs_key key;
Chris Masone02119d2008-09-05 16:13:11 -04001627 int level;
1628 int i;
1629 int ret;
1630
1631 btrfs_read_buffer(eb, gen);
1632
1633 level = btrfs_header_level(eb);
1634
1635 if (level != 0)
1636 return 0;
1637
1638 path = btrfs_alloc_path();
Mark Fasheh1e5063d2011-07-12 10:46:06 -07001639 if (!path)
1640 return -ENOMEM;
Chris Masone02119d2008-09-05 16:13:11 -04001641
1642 nritems = btrfs_header_nritems(eb);
1643 for (i = 0; i < nritems; i++) {
1644 btrfs_item_key_to_cpu(eb, &key, i);
Chris Masone02119d2008-09-05 16:13:11 -04001645
1646 /* inode keys are done during the first stage */
1647 if (key.type == BTRFS_INODE_ITEM_KEY &&
1648 wc->stage == LOG_WALK_REPLAY_INODES) {
Chris Masone02119d2008-09-05 16:13:11 -04001649 struct btrfs_inode_item *inode_item;
1650 u32 mode;
1651
1652 inode_item = btrfs_item_ptr(eb, i,
1653 struct btrfs_inode_item);
1654 mode = btrfs_inode_mode(eb, inode_item);
1655 if (S_ISDIR(mode)) {
1656 ret = replay_dir_deletes(wc->trans,
Chris Mason12fcfd22009-03-24 10:24:20 -04001657 root, log, path, key.objectid, 0);
Chris Masone02119d2008-09-05 16:13:11 -04001658 BUG_ON(ret);
1659 }
1660 ret = overwrite_item(wc->trans, root, path,
1661 eb, i, &key);
1662 BUG_ON(ret);
1663
Yan, Zhengc71bf092009-11-12 09:34:40 +00001664 /* for regular files, make sure corresponding
1665 * orhpan item exist. extents past the new EOF
1666 * will be truncated later by orphan cleanup.
Chris Masone02119d2008-09-05 16:13:11 -04001667 */
1668 if (S_ISREG(mode)) {
Yan, Zhengc71bf092009-11-12 09:34:40 +00001669 ret = insert_orphan_item(wc->trans, root,
1670 key.objectid);
Chris Masone02119d2008-09-05 16:13:11 -04001671 BUG_ON(ret);
Chris Masone02119d2008-09-05 16:13:11 -04001672 }
Yan, Zhengc71bf092009-11-12 09:34:40 +00001673
Chris Masone02119d2008-09-05 16:13:11 -04001674 ret = link_to_fixup_dir(wc->trans, root,
1675 path, key.objectid);
1676 BUG_ON(ret);
1677 }
1678 if (wc->stage < LOG_WALK_REPLAY_ALL)
1679 continue;
1680
1681 /* these keys are simply copied */
1682 if (key.type == BTRFS_XATTR_ITEM_KEY) {
1683 ret = overwrite_item(wc->trans, root, path,
1684 eb, i, &key);
1685 BUG_ON(ret);
1686 } else if (key.type == BTRFS_INODE_REF_KEY) {
1687 ret = add_inode_ref(wc->trans, root, log, path,
1688 eb, i, &key);
1689 BUG_ON(ret && ret != -ENOENT);
1690 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
1691 ret = replay_one_extent(wc->trans, root, path,
1692 eb, i, &key);
1693 BUG_ON(ret);
Chris Masone02119d2008-09-05 16:13:11 -04001694 } else if (key.type == BTRFS_DIR_ITEM_KEY ||
1695 key.type == BTRFS_DIR_INDEX_KEY) {
1696 ret = replay_one_dir_item(wc->trans, root, path,
1697 eb, i, &key);
1698 BUG_ON(ret);
1699 }
1700 }
1701 btrfs_free_path(path);
1702 return 0;
1703}
1704
Chris Masond3977122009-01-05 21:25:51 -05001705static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
Chris Masone02119d2008-09-05 16:13:11 -04001706 struct btrfs_root *root,
1707 struct btrfs_path *path, int *level,
1708 struct walk_control *wc)
1709{
1710 u64 root_owner;
Chris Masone02119d2008-09-05 16:13:11 -04001711 u64 bytenr;
1712 u64 ptr_gen;
1713 struct extent_buffer *next;
1714 struct extent_buffer *cur;
1715 struct extent_buffer *parent;
1716 u32 blocksize;
1717 int ret = 0;
1718
1719 WARN_ON(*level < 0);
1720 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1721
Chris Masond3977122009-01-05 21:25:51 -05001722 while (*level > 0) {
Chris Masone02119d2008-09-05 16:13:11 -04001723 WARN_ON(*level < 0);
1724 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1725 cur = path->nodes[*level];
1726
1727 if (btrfs_header_level(cur) != *level)
1728 WARN_ON(1);
1729
1730 if (path->slots[*level] >=
1731 btrfs_header_nritems(cur))
1732 break;
1733
1734 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1735 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
1736 blocksize = btrfs_level_size(root, *level - 1);
1737
1738 parent = path->nodes[*level];
1739 root_owner = btrfs_header_owner(parent);
Chris Masone02119d2008-09-05 16:13:11 -04001740
1741 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
liubo2a29edc2011-01-26 06:22:08 +00001742 if (!next)
1743 return -ENOMEM;
Chris Masone02119d2008-09-05 16:13:11 -04001744
Chris Masone02119d2008-09-05 16:13:11 -04001745 if (*level == 1) {
Mark Fasheh1e5063d2011-07-12 10:46:06 -07001746 ret = wc->process_func(root, next, wc, ptr_gen);
1747 if (ret)
1748 return ret;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04001749
Chris Masone02119d2008-09-05 16:13:11 -04001750 path->slots[*level]++;
1751 if (wc->free) {
1752 btrfs_read_buffer(next, ptr_gen);
1753
1754 btrfs_tree_lock(next);
Chris Masonb4ce94d2009-02-04 09:25:08 -05001755 btrfs_set_lock_blocking(next);
Chris Masonbd681512011-07-16 15:23:14 -04001756 clean_tree_block(trans, root, next);
Chris Masone02119d2008-09-05 16:13:11 -04001757 btrfs_wait_tree_block_writeback(next);
1758 btrfs_tree_unlock(next);
1759
Chris Masone02119d2008-09-05 16:13:11 -04001760 WARN_ON(root_owner !=
1761 BTRFS_TREE_LOG_OBJECTID);
Chris Masone688b7252011-10-31 20:52:39 -04001762 ret = btrfs_free_and_pin_reserved_extent(root,
Chris Masond00aff02008-09-11 15:54:42 -04001763 bytenr, blocksize);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001764 BUG_ON(ret); /* -ENOMEM or logic errors */
Chris Masone02119d2008-09-05 16:13:11 -04001765 }
1766 free_extent_buffer(next);
1767 continue;
1768 }
1769 btrfs_read_buffer(next, ptr_gen);
1770
1771 WARN_ON(*level <= 0);
1772 if (path->nodes[*level-1])
1773 free_extent_buffer(path->nodes[*level-1]);
1774 path->nodes[*level-1] = next;
1775 *level = btrfs_header_level(next);
1776 path->slots[*level] = 0;
1777 cond_resched();
1778 }
1779 WARN_ON(*level < 0);
1780 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1781
Yan, Zheng4a500fd2010-05-16 10:49:59 -04001782 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
Chris Masone02119d2008-09-05 16:13:11 -04001783
1784 cond_resched();
1785 return 0;
1786}
1787
Chris Masond3977122009-01-05 21:25:51 -05001788static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
Chris Masone02119d2008-09-05 16:13:11 -04001789 struct btrfs_root *root,
1790 struct btrfs_path *path, int *level,
1791 struct walk_control *wc)
1792{
1793 u64 root_owner;
Chris Masone02119d2008-09-05 16:13:11 -04001794 int i;
1795 int slot;
1796 int ret;
1797
Chris Masond3977122009-01-05 21:25:51 -05001798 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
Chris Masone02119d2008-09-05 16:13:11 -04001799 slot = path->slots[i];
Yan, Zheng4a500fd2010-05-16 10:49:59 -04001800 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
Chris Masone02119d2008-09-05 16:13:11 -04001801 path->slots[i]++;
1802 *level = i;
1803 WARN_ON(*level == 0);
1804 return 0;
1805 } else {
Zheng Yan31840ae2008-09-23 13:14:14 -04001806 struct extent_buffer *parent;
1807 if (path->nodes[*level] == root->node)
1808 parent = path->nodes[*level];
1809 else
1810 parent = path->nodes[*level + 1];
1811
1812 root_owner = btrfs_header_owner(parent);
Mark Fasheh1e5063d2011-07-12 10:46:06 -07001813 ret = wc->process_func(root, path->nodes[*level], wc,
Chris Masone02119d2008-09-05 16:13:11 -04001814 btrfs_header_generation(path->nodes[*level]));
Mark Fasheh1e5063d2011-07-12 10:46:06 -07001815 if (ret)
1816 return ret;
1817
Chris Masone02119d2008-09-05 16:13:11 -04001818 if (wc->free) {
1819 struct extent_buffer *next;
1820
1821 next = path->nodes[*level];
1822
1823 btrfs_tree_lock(next);
Chris Masonb4ce94d2009-02-04 09:25:08 -05001824 btrfs_set_lock_blocking(next);
Chris Masonbd681512011-07-16 15:23:14 -04001825 clean_tree_block(trans, root, next);
Chris Masone02119d2008-09-05 16:13:11 -04001826 btrfs_wait_tree_block_writeback(next);
1827 btrfs_tree_unlock(next);
1828
Chris Masone02119d2008-09-05 16:13:11 -04001829 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
Chris Masone688b7252011-10-31 20:52:39 -04001830 ret = btrfs_free_and_pin_reserved_extent(root,
Chris Masone02119d2008-09-05 16:13:11 -04001831 path->nodes[*level]->start,
Chris Masond00aff02008-09-11 15:54:42 -04001832 path->nodes[*level]->len);
Chris Masone02119d2008-09-05 16:13:11 -04001833 BUG_ON(ret);
1834 }
1835 free_extent_buffer(path->nodes[*level]);
1836 path->nodes[*level] = NULL;
1837 *level = i + 1;
1838 }
1839 }
1840 return 1;
1841}
1842
1843/*
1844 * drop the reference count on the tree rooted at 'snap'. This traverses
1845 * the tree freeing any blocks that have a ref count of zero after being
1846 * decremented.
1847 */
1848static int walk_log_tree(struct btrfs_trans_handle *trans,
1849 struct btrfs_root *log, struct walk_control *wc)
1850{
1851 int ret = 0;
1852 int wret;
1853 int level;
1854 struct btrfs_path *path;
1855 int i;
1856 int orig_level;
1857
1858 path = btrfs_alloc_path();
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00001859 if (!path)
1860 return -ENOMEM;
Chris Masone02119d2008-09-05 16:13:11 -04001861
1862 level = btrfs_header_level(log->node);
1863 orig_level = level;
1864 path->nodes[level] = log->node;
1865 extent_buffer_get(log->node);
1866 path->slots[level] = 0;
1867
Chris Masond3977122009-01-05 21:25:51 -05001868 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04001869 wret = walk_down_log_tree(trans, log, path, &level, wc);
1870 if (wret > 0)
1871 break;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001872 if (wret < 0) {
Chris Masone02119d2008-09-05 16:13:11 -04001873 ret = wret;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001874 goto out;
1875 }
Chris Masone02119d2008-09-05 16:13:11 -04001876
1877 wret = walk_up_log_tree(trans, log, path, &level, wc);
1878 if (wret > 0)
1879 break;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001880 if (wret < 0) {
Chris Masone02119d2008-09-05 16:13:11 -04001881 ret = wret;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001882 goto out;
1883 }
Chris Masone02119d2008-09-05 16:13:11 -04001884 }
1885
1886 /* was the root node processed? if not, catch it here */
1887 if (path->nodes[orig_level]) {
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001888 ret = wc->process_func(log, path->nodes[orig_level], wc,
Chris Masone02119d2008-09-05 16:13:11 -04001889 btrfs_header_generation(path->nodes[orig_level]));
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001890 if (ret)
1891 goto out;
Chris Masone02119d2008-09-05 16:13:11 -04001892 if (wc->free) {
1893 struct extent_buffer *next;
1894
1895 next = path->nodes[orig_level];
1896
1897 btrfs_tree_lock(next);
Chris Masonb4ce94d2009-02-04 09:25:08 -05001898 btrfs_set_lock_blocking(next);
Chris Masonbd681512011-07-16 15:23:14 -04001899 clean_tree_block(trans, log, next);
Chris Masone02119d2008-09-05 16:13:11 -04001900 btrfs_wait_tree_block_writeback(next);
1901 btrfs_tree_unlock(next);
1902
Chris Masone02119d2008-09-05 16:13:11 -04001903 WARN_ON(log->root_key.objectid !=
1904 BTRFS_TREE_LOG_OBJECTID);
Chris Masone688b7252011-10-31 20:52:39 -04001905 ret = btrfs_free_and_pin_reserved_extent(log, next->start,
Chris Masond00aff02008-09-11 15:54:42 -04001906 next->len);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001907 BUG_ON(ret); /* -ENOMEM or logic errors */
Chris Masone02119d2008-09-05 16:13:11 -04001908 }
1909 }
1910
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001911out:
Chris Masone02119d2008-09-05 16:13:11 -04001912 for (i = 0; i <= orig_level; i++) {
1913 if (path->nodes[i]) {
1914 free_extent_buffer(path->nodes[i]);
1915 path->nodes[i] = NULL;
1916 }
1917 }
1918 btrfs_free_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04001919 return ret;
1920}
1921
Yan Zheng7237f182009-01-21 12:54:03 -05001922/*
1923 * helper function to update the item for a given subvolumes log root
1924 * in the tree of log roots
1925 */
1926static int update_log_root(struct btrfs_trans_handle *trans,
1927 struct btrfs_root *log)
1928{
1929 int ret;
1930
1931 if (log->log_transid == 1) {
1932 /* insert root item on the first sync */
1933 ret = btrfs_insert_root(trans, log->fs_info->log_root_tree,
1934 &log->root_key, &log->root_item);
1935 } else {
1936 ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
1937 &log->root_key, &log->root_item);
1938 }
1939 return ret;
1940}
1941
Chris Mason12fcfd22009-03-24 10:24:20 -04001942static int wait_log_commit(struct btrfs_trans_handle *trans,
1943 struct btrfs_root *root, unsigned long transid)
Chris Masone02119d2008-09-05 16:13:11 -04001944{
1945 DEFINE_WAIT(wait);
Yan Zheng7237f182009-01-21 12:54:03 -05001946 int index = transid % 2;
Chris Masone02119d2008-09-05 16:13:11 -04001947
Yan Zheng7237f182009-01-21 12:54:03 -05001948 /*
1949 * we only allow two pending log transactions at a time,
1950 * so we know that if ours is more than 2 older than the
1951 * current transaction, we're done
1952 */
Chris Masone02119d2008-09-05 16:13:11 -04001953 do {
Yan Zheng7237f182009-01-21 12:54:03 -05001954 prepare_to_wait(&root->log_commit_wait[index],
1955 &wait, TASK_UNINTERRUPTIBLE);
1956 mutex_unlock(&root->log_mutex);
Chris Mason12fcfd22009-03-24 10:24:20 -04001957
1958 if (root->fs_info->last_trans_log_full_commit !=
1959 trans->transid && root->log_transid < transid + 2 &&
Yan Zheng7237f182009-01-21 12:54:03 -05001960 atomic_read(&root->log_commit[index]))
Chris Masone02119d2008-09-05 16:13:11 -04001961 schedule();
Chris Mason12fcfd22009-03-24 10:24:20 -04001962
Yan Zheng7237f182009-01-21 12:54:03 -05001963 finish_wait(&root->log_commit_wait[index], &wait);
1964 mutex_lock(&root->log_mutex);
Jan Kara6dd70ce2012-01-26 15:01:11 -05001965 } while (root->fs_info->last_trans_log_full_commit !=
1966 trans->transid && root->log_transid < transid + 2 &&
Yan Zheng7237f182009-01-21 12:54:03 -05001967 atomic_read(&root->log_commit[index]));
1968 return 0;
1969}
1970
Jeff Mahoney143bede2012-03-01 14:56:26 +01001971static void wait_for_writer(struct btrfs_trans_handle *trans,
1972 struct btrfs_root *root)
Yan Zheng7237f182009-01-21 12:54:03 -05001973{
1974 DEFINE_WAIT(wait);
Jan Kara6dd70ce2012-01-26 15:01:11 -05001975 while (root->fs_info->last_trans_log_full_commit !=
1976 trans->transid && atomic_read(&root->log_writers)) {
Yan Zheng7237f182009-01-21 12:54:03 -05001977 prepare_to_wait(&root->log_writer_wait,
1978 &wait, TASK_UNINTERRUPTIBLE);
1979 mutex_unlock(&root->log_mutex);
Chris Mason12fcfd22009-03-24 10:24:20 -04001980 if (root->fs_info->last_trans_log_full_commit !=
1981 trans->transid && atomic_read(&root->log_writers))
Yan Zheng7237f182009-01-21 12:54:03 -05001982 schedule();
1983 mutex_lock(&root->log_mutex);
1984 finish_wait(&root->log_writer_wait, &wait);
1985 }
Chris Masone02119d2008-09-05 16:13:11 -04001986}
1987
1988/*
1989 * btrfs_sync_log does sends a given tree log down to the disk and
1990 * updates the super blocks to record it. When this call is done,
Chris Mason12fcfd22009-03-24 10:24:20 -04001991 * you know that any inodes previously logged are safely on disk only
1992 * if it returns 0.
1993 *
1994 * Any other return value means you need to call btrfs_commit_transaction.
1995 * Some of the edge cases for fsyncing directories that have had unlinks
1996 * or renames done in the past mean that sometimes the only safe
1997 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
1998 * that has happened.
Chris Masone02119d2008-09-05 16:13:11 -04001999 */
2000int btrfs_sync_log(struct btrfs_trans_handle *trans,
2001 struct btrfs_root *root)
2002{
Yan Zheng7237f182009-01-21 12:54:03 -05002003 int index1;
2004 int index2;
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002005 int mark;
Chris Masone02119d2008-09-05 16:13:11 -04002006 int ret;
Chris Masone02119d2008-09-05 16:13:11 -04002007 struct btrfs_root *log = root->log_root;
Yan Zheng7237f182009-01-21 12:54:03 -05002008 struct btrfs_root *log_root_tree = root->fs_info->log_root_tree;
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002009 unsigned long log_transid = 0;
Chris Masone02119d2008-09-05 16:13:11 -04002010
Yan Zheng7237f182009-01-21 12:54:03 -05002011 mutex_lock(&root->log_mutex);
2012 index1 = root->log_transid % 2;
2013 if (atomic_read(&root->log_commit[index1])) {
Chris Mason12fcfd22009-03-24 10:24:20 -04002014 wait_log_commit(trans, root, root->log_transid);
Yan Zheng7237f182009-01-21 12:54:03 -05002015 mutex_unlock(&root->log_mutex);
2016 return 0;
Chris Masone02119d2008-09-05 16:13:11 -04002017 }
Yan Zheng7237f182009-01-21 12:54:03 -05002018 atomic_set(&root->log_commit[index1], 1);
2019
2020 /* wait for previous tree log sync to complete */
2021 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
Chris Mason12fcfd22009-03-24 10:24:20 -04002022 wait_log_commit(trans, root, root->log_transid - 1);
Yan, Zheng86df7eb2009-10-14 09:24:59 -04002023 while (1) {
Yan Zheng7237f182009-01-21 12:54:03 -05002024 unsigned long batch = root->log_batch;
Chris Masoncd354ad2011-10-20 15:45:37 -04002025 /* when we're on an ssd, just kick the log commit out */
2026 if (!btrfs_test_opt(root, SSD) && root->log_multiple_pids) {
Yan, Zheng86df7eb2009-10-14 09:24:59 -04002027 mutex_unlock(&root->log_mutex);
2028 schedule_timeout_uninterruptible(1);
2029 mutex_lock(&root->log_mutex);
2030 }
Chris Mason12fcfd22009-03-24 10:24:20 -04002031 wait_for_writer(trans, root);
Yan Zheng7237f182009-01-21 12:54:03 -05002032 if (batch == root->log_batch)
Chris Masone02119d2008-09-05 16:13:11 -04002033 break;
2034 }
Chris Masond0c803c2008-09-11 16:17:57 -04002035
Chris Mason12fcfd22009-03-24 10:24:20 -04002036 /* bail out if we need to do a full commit */
2037 if (root->fs_info->last_trans_log_full_commit == trans->transid) {
2038 ret = -EAGAIN;
2039 mutex_unlock(&root->log_mutex);
2040 goto out;
2041 }
2042
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002043 log_transid = root->log_transid;
2044 if (log_transid % 2 == 0)
2045 mark = EXTENT_DIRTY;
2046 else
2047 mark = EXTENT_NEW;
2048
Chris Mason690587d2009-10-13 13:29:19 -04002049 /* we start IO on all the marked extents here, but we don't actually
2050 * wait for them until later.
2051 */
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002052 ret = btrfs_write_marked_extents(log, &log->dirty_log_pages, mark);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002053 if (ret) {
2054 btrfs_abort_transaction(trans, root, ret);
2055 mutex_unlock(&root->log_mutex);
2056 goto out;
2057 }
Yan Zheng7237f182009-01-21 12:54:03 -05002058
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002059 btrfs_set_root_node(&log->root_item, log->node);
Yan Zheng7237f182009-01-21 12:54:03 -05002060
2061 root->log_batch = 0;
2062 root->log_transid++;
2063 log->log_transid = root->log_transid;
Josef Bacikff782e02009-10-08 15:30:04 -04002064 root->log_start_pid = 0;
Yan Zheng7237f182009-01-21 12:54:03 -05002065 smp_mb();
2066 /*
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002067 * IO has been started, blocks of the log tree have WRITTEN flag set
2068 * in their headers. new modifications of the log will be written to
2069 * new positions. so it's safe to allow log writers to go in.
Yan Zheng7237f182009-01-21 12:54:03 -05002070 */
2071 mutex_unlock(&root->log_mutex);
2072
2073 mutex_lock(&log_root_tree->log_mutex);
2074 log_root_tree->log_batch++;
2075 atomic_inc(&log_root_tree->log_writers);
2076 mutex_unlock(&log_root_tree->log_mutex);
2077
2078 ret = update_log_root(trans, log);
Yan Zheng7237f182009-01-21 12:54:03 -05002079
2080 mutex_lock(&log_root_tree->log_mutex);
2081 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
2082 smp_mb();
2083 if (waitqueue_active(&log_root_tree->log_writer_wait))
2084 wake_up(&log_root_tree->log_writer_wait);
2085 }
2086
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002087 if (ret) {
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002088 if (ret != -ENOSPC) {
2089 btrfs_abort_transaction(trans, root, ret);
2090 mutex_unlock(&log_root_tree->log_mutex);
2091 goto out;
2092 }
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002093 root->fs_info->last_trans_log_full_commit = trans->transid;
2094 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2095 mutex_unlock(&log_root_tree->log_mutex);
2096 ret = -EAGAIN;
2097 goto out;
2098 }
2099
Yan Zheng7237f182009-01-21 12:54:03 -05002100 index2 = log_root_tree->log_transid % 2;
2101 if (atomic_read(&log_root_tree->log_commit[index2])) {
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002102 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
Chris Mason12fcfd22009-03-24 10:24:20 -04002103 wait_log_commit(trans, log_root_tree,
2104 log_root_tree->log_transid);
Yan Zheng7237f182009-01-21 12:54:03 -05002105 mutex_unlock(&log_root_tree->log_mutex);
Chris Masonb31eabd2011-01-31 16:48:24 -05002106 ret = 0;
Yan Zheng7237f182009-01-21 12:54:03 -05002107 goto out;
2108 }
2109 atomic_set(&log_root_tree->log_commit[index2], 1);
2110
Chris Mason12fcfd22009-03-24 10:24:20 -04002111 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
2112 wait_log_commit(trans, log_root_tree,
2113 log_root_tree->log_transid - 1);
2114 }
Yan Zheng7237f182009-01-21 12:54:03 -05002115
Chris Mason12fcfd22009-03-24 10:24:20 -04002116 wait_for_writer(trans, log_root_tree);
2117
2118 /*
2119 * now that we've moved on to the tree of log tree roots,
2120 * check the full commit flag again
2121 */
2122 if (root->fs_info->last_trans_log_full_commit == trans->transid) {
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002123 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
Chris Mason12fcfd22009-03-24 10:24:20 -04002124 mutex_unlock(&log_root_tree->log_mutex);
2125 ret = -EAGAIN;
2126 goto out_wake_log_root;
2127 }
Yan Zheng7237f182009-01-21 12:54:03 -05002128
2129 ret = btrfs_write_and_wait_marked_extents(log_root_tree,
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002130 &log_root_tree->dirty_log_pages,
2131 EXTENT_DIRTY | EXTENT_NEW);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002132 if (ret) {
2133 btrfs_abort_transaction(trans, root, ret);
2134 mutex_unlock(&log_root_tree->log_mutex);
2135 goto out_wake_log_root;
2136 }
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002137 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
Chris Masone02119d2008-09-05 16:13:11 -04002138
David Sterba6c417612011-04-13 15:41:04 +02002139 btrfs_set_super_log_root(root->fs_info->super_for_commit,
Yan Zheng7237f182009-01-21 12:54:03 -05002140 log_root_tree->node->start);
David Sterba6c417612011-04-13 15:41:04 +02002141 btrfs_set_super_log_root_level(root->fs_info->super_for_commit,
Yan Zheng7237f182009-01-21 12:54:03 -05002142 btrfs_header_level(log_root_tree->node));
Chris Masone02119d2008-09-05 16:13:11 -04002143
Yan Zheng7237f182009-01-21 12:54:03 -05002144 log_root_tree->log_batch = 0;
2145 log_root_tree->log_transid++;
Chris Masone02119d2008-09-05 16:13:11 -04002146 smp_mb();
Yan Zheng7237f182009-01-21 12:54:03 -05002147
2148 mutex_unlock(&log_root_tree->log_mutex);
2149
2150 /*
2151 * nobody else is going to jump in and write the the ctree
2152 * super here because the log_commit atomic below is protecting
2153 * us. We must be called with a transaction handle pinning
2154 * the running transaction open, so a full commit can't hop
2155 * in and cause problems either.
2156 */
Arne Jansena2de7332011-03-08 14:14:00 +01002157 btrfs_scrub_pause_super(root);
Chris Mason47226072009-10-13 12:55:09 -04002158 write_ctree_super(trans, root->fs_info->tree_root, 1);
Arne Jansena2de7332011-03-08 14:14:00 +01002159 btrfs_scrub_continue_super(root);
Chris Mason12fcfd22009-03-24 10:24:20 -04002160 ret = 0;
Yan Zheng7237f182009-01-21 12:54:03 -05002161
Chris Mason257c62e2009-10-13 13:21:08 -04002162 mutex_lock(&root->log_mutex);
2163 if (root->last_log_commit < log_transid)
2164 root->last_log_commit = log_transid;
2165 mutex_unlock(&root->log_mutex);
2166
Chris Mason12fcfd22009-03-24 10:24:20 -04002167out_wake_log_root:
Yan Zheng7237f182009-01-21 12:54:03 -05002168 atomic_set(&log_root_tree->log_commit[index2], 0);
2169 smp_mb();
2170 if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
2171 wake_up(&log_root_tree->log_commit_wait[index2]);
Chris Masone02119d2008-09-05 16:13:11 -04002172out:
Yan Zheng7237f182009-01-21 12:54:03 -05002173 atomic_set(&root->log_commit[index1], 0);
2174 smp_mb();
2175 if (waitqueue_active(&root->log_commit_wait[index1]))
2176 wake_up(&root->log_commit_wait[index1]);
Chris Masonb31eabd2011-01-31 16:48:24 -05002177 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002178}
2179
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002180static void free_log_tree(struct btrfs_trans_handle *trans,
2181 struct btrfs_root *log)
Chris Masone02119d2008-09-05 16:13:11 -04002182{
2183 int ret;
Chris Masond0c803c2008-09-11 16:17:57 -04002184 u64 start;
2185 u64 end;
Chris Masone02119d2008-09-05 16:13:11 -04002186 struct walk_control wc = {
2187 .free = 1,
2188 .process_func = process_one_buffer
2189 };
2190
Chris Masone02119d2008-09-05 16:13:11 -04002191 ret = walk_log_tree(trans, log, &wc);
2192 BUG_ON(ret);
2193
Chris Masond3977122009-01-05 21:25:51 -05002194 while (1) {
Chris Masond0c803c2008-09-11 16:17:57 -04002195 ret = find_first_extent_bit(&log->dirty_log_pages,
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002196 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW);
Chris Masond0c803c2008-09-11 16:17:57 -04002197 if (ret)
2198 break;
2199
Yan, Zheng8cef4e12009-11-12 09:33:26 +00002200 clear_extent_bits(&log->dirty_log_pages, start, end,
2201 EXTENT_DIRTY | EXTENT_NEW, GFP_NOFS);
Chris Masond0c803c2008-09-11 16:17:57 -04002202 }
2203
Yan Zheng7237f182009-01-21 12:54:03 -05002204 free_extent_buffer(log->node);
2205 kfree(log);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002206}
2207
2208/*
2209 * free all the extents used by the tree log. This should be called
2210 * at commit time of the full transaction
2211 */
2212int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
2213{
2214 if (root->log_root) {
2215 free_log_tree(trans, root->log_root);
2216 root->log_root = NULL;
2217 }
2218 return 0;
2219}
2220
2221int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
2222 struct btrfs_fs_info *fs_info)
2223{
2224 if (fs_info->log_root_tree) {
2225 free_log_tree(trans, fs_info->log_root_tree);
2226 fs_info->log_root_tree = NULL;
2227 }
Chris Masone02119d2008-09-05 16:13:11 -04002228 return 0;
2229}
2230
2231/*
Chris Masone02119d2008-09-05 16:13:11 -04002232 * If both a file and directory are logged, and unlinks or renames are
2233 * mixed in, we have a few interesting corners:
2234 *
2235 * create file X in dir Y
2236 * link file X to X.link in dir Y
2237 * fsync file X
2238 * unlink file X but leave X.link
2239 * fsync dir Y
2240 *
2241 * After a crash we would expect only X.link to exist. But file X
2242 * didn't get fsync'd again so the log has back refs for X and X.link.
2243 *
2244 * We solve this by removing directory entries and inode backrefs from the
2245 * log when a file that was logged in the current transaction is
2246 * unlinked. Any later fsync will include the updated log entries, and
2247 * we'll be able to reconstruct the proper directory items from backrefs.
2248 *
2249 * This optimizations allows us to avoid relogging the entire inode
2250 * or the entire directory.
2251 */
2252int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
2253 struct btrfs_root *root,
2254 const char *name, int name_len,
2255 struct inode *dir, u64 index)
2256{
2257 struct btrfs_root *log;
2258 struct btrfs_dir_item *di;
2259 struct btrfs_path *path;
2260 int ret;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002261 int err = 0;
Chris Masone02119d2008-09-05 16:13:11 -04002262 int bytes_del = 0;
Li Zefan33345d012011-04-20 10:31:50 +08002263 u64 dir_ino = btrfs_ino(dir);
Chris Masone02119d2008-09-05 16:13:11 -04002264
Chris Mason3a5f1d42008-09-11 15:53:37 -04002265 if (BTRFS_I(dir)->logged_trans < trans->transid)
2266 return 0;
2267
Chris Masone02119d2008-09-05 16:13:11 -04002268 ret = join_running_log_trans(root);
2269 if (ret)
2270 return 0;
2271
2272 mutex_lock(&BTRFS_I(dir)->log_mutex);
2273
2274 log = root->log_root;
2275 path = btrfs_alloc_path();
Tsutomu Itoha62f44a2011-04-25 19:43:51 -04002276 if (!path) {
2277 err = -ENOMEM;
2278 goto out_unlock;
2279 }
liubo2a29edc2011-01-26 06:22:08 +00002280
Li Zefan33345d012011-04-20 10:31:50 +08002281 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
Chris Masone02119d2008-09-05 16:13:11 -04002282 name, name_len, -1);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002283 if (IS_ERR(di)) {
2284 err = PTR_ERR(di);
2285 goto fail;
2286 }
2287 if (di) {
Chris Masone02119d2008-09-05 16:13:11 -04002288 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2289 bytes_del += name_len;
2290 BUG_ON(ret);
2291 }
David Sterbab3b4aa72011-04-21 01:20:15 +02002292 btrfs_release_path(path);
Li Zefan33345d012011-04-20 10:31:50 +08002293 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
Chris Masone02119d2008-09-05 16:13:11 -04002294 index, name, name_len, -1);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002295 if (IS_ERR(di)) {
2296 err = PTR_ERR(di);
2297 goto fail;
2298 }
2299 if (di) {
Chris Masone02119d2008-09-05 16:13:11 -04002300 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2301 bytes_del += name_len;
2302 BUG_ON(ret);
2303 }
2304
2305 /* update the directory size in the log to reflect the names
2306 * we have removed
2307 */
2308 if (bytes_del) {
2309 struct btrfs_key key;
2310
Li Zefan33345d012011-04-20 10:31:50 +08002311 key.objectid = dir_ino;
Chris Masone02119d2008-09-05 16:13:11 -04002312 key.offset = 0;
2313 key.type = BTRFS_INODE_ITEM_KEY;
David Sterbab3b4aa72011-04-21 01:20:15 +02002314 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002315
2316 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002317 if (ret < 0) {
2318 err = ret;
2319 goto fail;
2320 }
Chris Masone02119d2008-09-05 16:13:11 -04002321 if (ret == 0) {
2322 struct btrfs_inode_item *item;
2323 u64 i_size;
2324
2325 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2326 struct btrfs_inode_item);
2327 i_size = btrfs_inode_size(path->nodes[0], item);
2328 if (i_size > bytes_del)
2329 i_size -= bytes_del;
2330 else
2331 i_size = 0;
2332 btrfs_set_inode_size(path->nodes[0], item, i_size);
2333 btrfs_mark_buffer_dirty(path->nodes[0]);
2334 } else
2335 ret = 0;
David Sterbab3b4aa72011-04-21 01:20:15 +02002336 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002337 }
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002338fail:
Chris Masone02119d2008-09-05 16:13:11 -04002339 btrfs_free_path(path);
Tsutomu Itoha62f44a2011-04-25 19:43:51 -04002340out_unlock:
Chris Masone02119d2008-09-05 16:13:11 -04002341 mutex_unlock(&BTRFS_I(dir)->log_mutex);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002342 if (ret == -ENOSPC) {
2343 root->fs_info->last_trans_log_full_commit = trans->transid;
2344 ret = 0;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002345 } else if (ret < 0)
2346 btrfs_abort_transaction(trans, root, ret);
2347
Chris Mason12fcfd22009-03-24 10:24:20 -04002348 btrfs_end_log_trans(root);
Chris Masone02119d2008-09-05 16:13:11 -04002349
Andi Kleen411fc6b2010-10-29 15:14:31 -04002350 return err;
Chris Masone02119d2008-09-05 16:13:11 -04002351}
2352
2353/* see comments for btrfs_del_dir_entries_in_log */
2354int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
2355 struct btrfs_root *root,
2356 const char *name, int name_len,
2357 struct inode *inode, u64 dirid)
2358{
2359 struct btrfs_root *log;
2360 u64 index;
2361 int ret;
2362
Chris Mason3a5f1d42008-09-11 15:53:37 -04002363 if (BTRFS_I(inode)->logged_trans < trans->transid)
2364 return 0;
2365
Chris Masone02119d2008-09-05 16:13:11 -04002366 ret = join_running_log_trans(root);
2367 if (ret)
2368 return 0;
2369 log = root->log_root;
2370 mutex_lock(&BTRFS_I(inode)->log_mutex);
2371
Li Zefan33345d012011-04-20 10:31:50 +08002372 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
Chris Masone02119d2008-09-05 16:13:11 -04002373 dirid, &index);
2374 mutex_unlock(&BTRFS_I(inode)->log_mutex);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002375 if (ret == -ENOSPC) {
2376 root->fs_info->last_trans_log_full_commit = trans->transid;
2377 ret = 0;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002378 } else if (ret < 0 && ret != -ENOENT)
2379 btrfs_abort_transaction(trans, root, ret);
Chris Mason12fcfd22009-03-24 10:24:20 -04002380 btrfs_end_log_trans(root);
Chris Masone02119d2008-09-05 16:13:11 -04002381
Chris Masone02119d2008-09-05 16:13:11 -04002382 return ret;
2383}
2384
2385/*
2386 * creates a range item in the log for 'dirid'. first_offset and
2387 * last_offset tell us which parts of the key space the log should
2388 * be considered authoritative for.
2389 */
2390static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
2391 struct btrfs_root *log,
2392 struct btrfs_path *path,
2393 int key_type, u64 dirid,
2394 u64 first_offset, u64 last_offset)
2395{
2396 int ret;
2397 struct btrfs_key key;
2398 struct btrfs_dir_log_item *item;
2399
2400 key.objectid = dirid;
2401 key.offset = first_offset;
2402 if (key_type == BTRFS_DIR_ITEM_KEY)
2403 key.type = BTRFS_DIR_LOG_ITEM_KEY;
2404 else
2405 key.type = BTRFS_DIR_LOG_INDEX_KEY;
2406 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002407 if (ret)
2408 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002409
2410 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2411 struct btrfs_dir_log_item);
2412 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
2413 btrfs_mark_buffer_dirty(path->nodes[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +02002414 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002415 return 0;
2416}
2417
2418/*
2419 * log all the items included in the current transaction for a given
2420 * directory. This also creates the range items in the log tree required
2421 * to replay anything deleted before the fsync
2422 */
2423static noinline int log_dir_items(struct btrfs_trans_handle *trans,
2424 struct btrfs_root *root, struct inode *inode,
2425 struct btrfs_path *path,
2426 struct btrfs_path *dst_path, int key_type,
2427 u64 min_offset, u64 *last_offset_ret)
2428{
2429 struct btrfs_key min_key;
2430 struct btrfs_key max_key;
2431 struct btrfs_root *log = root->log_root;
2432 struct extent_buffer *src;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002433 int err = 0;
Chris Masone02119d2008-09-05 16:13:11 -04002434 int ret;
2435 int i;
2436 int nritems;
2437 u64 first_offset = min_offset;
2438 u64 last_offset = (u64)-1;
Li Zefan33345d012011-04-20 10:31:50 +08002439 u64 ino = btrfs_ino(inode);
Chris Masone02119d2008-09-05 16:13:11 -04002440
2441 log = root->log_root;
Li Zefan33345d012011-04-20 10:31:50 +08002442 max_key.objectid = ino;
Chris Masone02119d2008-09-05 16:13:11 -04002443 max_key.offset = (u64)-1;
2444 max_key.type = key_type;
2445
Li Zefan33345d012011-04-20 10:31:50 +08002446 min_key.objectid = ino;
Chris Masone02119d2008-09-05 16:13:11 -04002447 min_key.type = key_type;
2448 min_key.offset = min_offset;
2449
2450 path->keep_locks = 1;
2451
2452 ret = btrfs_search_forward(root, &min_key, &max_key,
2453 path, 0, trans->transid);
2454
2455 /*
2456 * we didn't find anything from this transaction, see if there
2457 * is anything at all
2458 */
Li Zefan33345d012011-04-20 10:31:50 +08002459 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
2460 min_key.objectid = ino;
Chris Masone02119d2008-09-05 16:13:11 -04002461 min_key.type = key_type;
2462 min_key.offset = (u64)-1;
David Sterbab3b4aa72011-04-21 01:20:15 +02002463 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002464 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2465 if (ret < 0) {
David Sterbab3b4aa72011-04-21 01:20:15 +02002466 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002467 return ret;
2468 }
Li Zefan33345d012011-04-20 10:31:50 +08002469 ret = btrfs_previous_item(root, path, ino, key_type);
Chris Masone02119d2008-09-05 16:13:11 -04002470
2471 /* if ret == 0 there are items for this type,
2472 * create a range to tell us the last key of this type.
2473 * otherwise, there are no items in this directory after
2474 * *min_offset, and we create a range to indicate that.
2475 */
2476 if (ret == 0) {
2477 struct btrfs_key tmp;
2478 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
2479 path->slots[0]);
Chris Masond3977122009-01-05 21:25:51 -05002480 if (key_type == tmp.type)
Chris Masone02119d2008-09-05 16:13:11 -04002481 first_offset = max(min_offset, tmp.offset) + 1;
Chris Masone02119d2008-09-05 16:13:11 -04002482 }
2483 goto done;
2484 }
2485
2486 /* go backward to find any previous key */
Li Zefan33345d012011-04-20 10:31:50 +08002487 ret = btrfs_previous_item(root, path, ino, key_type);
Chris Masone02119d2008-09-05 16:13:11 -04002488 if (ret == 0) {
2489 struct btrfs_key tmp;
2490 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2491 if (key_type == tmp.type) {
2492 first_offset = tmp.offset;
2493 ret = overwrite_item(trans, log, dst_path,
2494 path->nodes[0], path->slots[0],
2495 &tmp);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002496 if (ret) {
2497 err = ret;
2498 goto done;
2499 }
Chris Masone02119d2008-09-05 16:13:11 -04002500 }
2501 }
David Sterbab3b4aa72011-04-21 01:20:15 +02002502 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002503
2504 /* find the first key from this transaction again */
2505 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2506 if (ret != 0) {
2507 WARN_ON(1);
2508 goto done;
2509 }
2510
2511 /*
2512 * we have a block from this transaction, log every item in it
2513 * from our directory
2514 */
Chris Masond3977122009-01-05 21:25:51 -05002515 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04002516 struct btrfs_key tmp;
2517 src = path->nodes[0];
2518 nritems = btrfs_header_nritems(src);
2519 for (i = path->slots[0]; i < nritems; i++) {
2520 btrfs_item_key_to_cpu(src, &min_key, i);
2521
Li Zefan33345d012011-04-20 10:31:50 +08002522 if (min_key.objectid != ino || min_key.type != key_type)
Chris Masone02119d2008-09-05 16:13:11 -04002523 goto done;
2524 ret = overwrite_item(trans, log, dst_path, src, i,
2525 &min_key);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002526 if (ret) {
2527 err = ret;
2528 goto done;
2529 }
Chris Masone02119d2008-09-05 16:13:11 -04002530 }
2531 path->slots[0] = nritems;
2532
2533 /*
2534 * look ahead to the next item and see if it is also
2535 * from this directory and from this transaction
2536 */
2537 ret = btrfs_next_leaf(root, path);
2538 if (ret == 1) {
2539 last_offset = (u64)-1;
2540 goto done;
2541 }
2542 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
Li Zefan33345d012011-04-20 10:31:50 +08002543 if (tmp.objectid != ino || tmp.type != key_type) {
Chris Masone02119d2008-09-05 16:13:11 -04002544 last_offset = (u64)-1;
2545 goto done;
2546 }
2547 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
2548 ret = overwrite_item(trans, log, dst_path,
2549 path->nodes[0], path->slots[0],
2550 &tmp);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002551 if (ret)
2552 err = ret;
2553 else
2554 last_offset = tmp.offset;
Chris Masone02119d2008-09-05 16:13:11 -04002555 goto done;
2556 }
2557 }
2558done:
David Sterbab3b4aa72011-04-21 01:20:15 +02002559 btrfs_release_path(path);
2560 btrfs_release_path(dst_path);
Chris Masone02119d2008-09-05 16:13:11 -04002561
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002562 if (err == 0) {
2563 *last_offset_ret = last_offset;
2564 /*
2565 * insert the log range keys to indicate where the log
2566 * is valid
2567 */
2568 ret = insert_dir_log_key(trans, log, path, key_type,
Li Zefan33345d012011-04-20 10:31:50 +08002569 ino, first_offset, last_offset);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002570 if (ret)
2571 err = ret;
2572 }
2573 return err;
Chris Masone02119d2008-09-05 16:13:11 -04002574}
2575
2576/*
2577 * logging directories is very similar to logging inodes, We find all the items
2578 * from the current transaction and write them to the log.
2579 *
2580 * The recovery code scans the directory in the subvolume, and if it finds a
2581 * key in the range logged that is not present in the log tree, then it means
2582 * that dir entry was unlinked during the transaction.
2583 *
2584 * In order for that scan to work, we must include one key smaller than
2585 * the smallest logged by this transaction and one key larger than the largest
2586 * key logged by this transaction.
2587 */
2588static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
2589 struct btrfs_root *root, struct inode *inode,
2590 struct btrfs_path *path,
2591 struct btrfs_path *dst_path)
2592{
2593 u64 min_key;
2594 u64 max_key;
2595 int ret;
2596 int key_type = BTRFS_DIR_ITEM_KEY;
2597
2598again:
2599 min_key = 0;
2600 max_key = 0;
Chris Masond3977122009-01-05 21:25:51 -05002601 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04002602 ret = log_dir_items(trans, root, inode, path,
2603 dst_path, key_type, min_key,
2604 &max_key);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002605 if (ret)
2606 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002607 if (max_key == (u64)-1)
2608 break;
2609 min_key = max_key + 1;
2610 }
2611
2612 if (key_type == BTRFS_DIR_ITEM_KEY) {
2613 key_type = BTRFS_DIR_INDEX_KEY;
2614 goto again;
2615 }
2616 return 0;
2617}
2618
2619/*
2620 * a helper function to drop items from the log before we relog an
2621 * inode. max_key_type indicates the highest item type to remove.
2622 * This cannot be run for file data extents because it does not
2623 * free the extents they point to.
2624 */
2625static int drop_objectid_items(struct btrfs_trans_handle *trans,
2626 struct btrfs_root *log,
2627 struct btrfs_path *path,
2628 u64 objectid, int max_key_type)
2629{
2630 int ret;
2631 struct btrfs_key key;
2632 struct btrfs_key found_key;
2633
2634 key.objectid = objectid;
2635 key.type = max_key_type;
2636 key.offset = (u64)-1;
2637
Chris Masond3977122009-01-05 21:25:51 -05002638 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04002639 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002640 BUG_ON(ret == 0);
2641 if (ret < 0)
Chris Masone02119d2008-09-05 16:13:11 -04002642 break;
2643
2644 if (path->slots[0] == 0)
2645 break;
2646
2647 path->slots[0]--;
2648 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2649 path->slots[0]);
2650
2651 if (found_key.objectid != objectid)
2652 break;
2653
2654 ret = btrfs_del_item(trans, log, path);
Tsutomu Itoh65a246c2011-05-19 04:37:44 +00002655 if (ret)
2656 break;
David Sterbab3b4aa72011-04-21 01:20:15 +02002657 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04002658 }
David Sterbab3b4aa72011-04-21 01:20:15 +02002659 btrfs_release_path(path);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002660 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04002661}
2662
Chris Mason31ff1cd2008-09-11 16:17:57 -04002663static noinline int copy_items(struct btrfs_trans_handle *trans,
2664 struct btrfs_root *log,
2665 struct btrfs_path *dst_path,
2666 struct extent_buffer *src,
2667 int start_slot, int nr, int inode_only)
2668{
2669 unsigned long src_offset;
2670 unsigned long dst_offset;
2671 struct btrfs_file_extent_item *extent;
2672 struct btrfs_inode_item *inode_item;
2673 int ret;
2674 struct btrfs_key *ins_keys;
2675 u32 *ins_sizes;
2676 char *ins_data;
2677 int i;
Chris Masond20f7042008-12-08 16:58:54 -05002678 struct list_head ordered_sums;
2679
2680 INIT_LIST_HEAD(&ordered_sums);
Chris Mason31ff1cd2008-09-11 16:17:57 -04002681
2682 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
2683 nr * sizeof(u32), GFP_NOFS);
liubo2a29edc2011-01-26 06:22:08 +00002684 if (!ins_data)
2685 return -ENOMEM;
2686
Chris Mason31ff1cd2008-09-11 16:17:57 -04002687 ins_sizes = (u32 *)ins_data;
2688 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
2689
2690 for (i = 0; i < nr; i++) {
2691 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
2692 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
2693 }
2694 ret = btrfs_insert_empty_items(trans, log, dst_path,
2695 ins_keys, ins_sizes, nr);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002696 if (ret) {
2697 kfree(ins_data);
2698 return ret;
2699 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04002700
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002701 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
Chris Mason31ff1cd2008-09-11 16:17:57 -04002702 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
2703 dst_path->slots[0]);
2704
2705 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
2706
2707 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
2708 src_offset, ins_sizes[i]);
2709
2710 if (inode_only == LOG_INODE_EXISTS &&
2711 ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
2712 inode_item = btrfs_item_ptr(dst_path->nodes[0],
2713 dst_path->slots[0],
2714 struct btrfs_inode_item);
2715 btrfs_set_inode_size(dst_path->nodes[0], inode_item, 0);
2716
2717 /* set the generation to zero so the recover code
2718 * can tell the difference between an logging
2719 * just to say 'this inode exists' and a logging
2720 * to say 'update this inode with these values'
2721 */
2722 btrfs_set_inode_generation(dst_path->nodes[0],
2723 inode_item, 0);
2724 }
2725 /* take a reference on file data extents so that truncates
2726 * or deletes of this inode don't have to relog the inode
2727 * again
2728 */
2729 if (btrfs_key_type(ins_keys + i) == BTRFS_EXTENT_DATA_KEY) {
2730 int found_type;
2731 extent = btrfs_item_ptr(src, start_slot + i,
2732 struct btrfs_file_extent_item);
2733
liubo8e531cd2011-05-06 10:36:09 +08002734 if (btrfs_file_extent_generation(src, extent) < trans->transid)
2735 continue;
2736
Chris Mason31ff1cd2008-09-11 16:17:57 -04002737 found_type = btrfs_file_extent_type(src, extent);
Yan Zhengd899e052008-10-30 14:25:28 -04002738 if (found_type == BTRFS_FILE_EXTENT_REG ||
2739 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002740 u64 ds, dl, cs, cl;
2741 ds = btrfs_file_extent_disk_bytenr(src,
2742 extent);
2743 /* ds == 0 is a hole */
2744 if (ds == 0)
2745 continue;
2746
2747 dl = btrfs_file_extent_disk_num_bytes(src,
2748 extent);
2749 cs = btrfs_file_extent_offset(src, extent);
2750 cl = btrfs_file_extent_num_bytes(src,
Joe Perchesa419aef2009-08-18 11:18:35 -07002751 extent);
Chris Mason580afd72008-12-08 19:15:39 -05002752 if (btrfs_file_extent_compression(src,
2753 extent)) {
2754 cs = 0;
2755 cl = dl;
2756 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002757
2758 ret = btrfs_lookup_csums_range(
2759 log->fs_info->csum_root,
2760 ds + cs, ds + cs + cl - 1,
Arne Jansena2de7332011-03-08 14:14:00 +01002761 &ordered_sums, 0);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002762 BUG_ON(ret);
Chris Mason31ff1cd2008-09-11 16:17:57 -04002763 }
2764 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04002765 }
2766
2767 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +02002768 btrfs_release_path(dst_path);
Chris Mason31ff1cd2008-09-11 16:17:57 -04002769 kfree(ins_data);
Chris Masond20f7042008-12-08 16:58:54 -05002770
2771 /*
2772 * we have to do this after the loop above to avoid changing the
2773 * log tree while trying to change the log tree.
2774 */
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002775 ret = 0;
Chris Masond3977122009-01-05 21:25:51 -05002776 while (!list_empty(&ordered_sums)) {
Chris Masond20f7042008-12-08 16:58:54 -05002777 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
2778 struct btrfs_ordered_sum,
2779 list);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002780 if (!ret)
2781 ret = btrfs_csum_file_blocks(trans, log, sums);
Chris Masond20f7042008-12-08 16:58:54 -05002782 list_del(&sums->list);
2783 kfree(sums);
2784 }
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002785 return ret;
Chris Mason31ff1cd2008-09-11 16:17:57 -04002786}
2787
Chris Masone02119d2008-09-05 16:13:11 -04002788/* log a single inode in the tree log.
2789 * At least one parent directory for this inode must exist in the tree
2790 * or be logged already.
2791 *
2792 * Any items from this inode changed by the current transaction are copied
2793 * to the log tree. An extra reference is taken on any extents in this
2794 * file, allowing us to avoid a whole pile of corner cases around logging
2795 * blocks that have been removed from the tree.
2796 *
2797 * See LOG_INODE_ALL and related defines for a description of what inode_only
2798 * does.
2799 *
2800 * This handles both files and directories.
2801 */
Chris Mason12fcfd22009-03-24 10:24:20 -04002802static int btrfs_log_inode(struct btrfs_trans_handle *trans,
Chris Masone02119d2008-09-05 16:13:11 -04002803 struct btrfs_root *root, struct inode *inode,
2804 int inode_only)
2805{
2806 struct btrfs_path *path;
2807 struct btrfs_path *dst_path;
2808 struct btrfs_key min_key;
2809 struct btrfs_key max_key;
2810 struct btrfs_root *log = root->log_root;
Chris Mason31ff1cd2008-09-11 16:17:57 -04002811 struct extent_buffer *src = NULL;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002812 int err = 0;
Chris Masone02119d2008-09-05 16:13:11 -04002813 int ret;
Chris Mason3a5f1d42008-09-11 15:53:37 -04002814 int nritems;
Chris Mason31ff1cd2008-09-11 16:17:57 -04002815 int ins_start_slot = 0;
2816 int ins_nr;
Li Zefan33345d012011-04-20 10:31:50 +08002817 u64 ino = btrfs_ino(inode);
Chris Masone02119d2008-09-05 16:13:11 -04002818
2819 log = root->log_root;
2820
2821 path = btrfs_alloc_path();
Tsutomu Itoh5df67082011-02-01 09:17:35 +00002822 if (!path)
2823 return -ENOMEM;
Chris Masone02119d2008-09-05 16:13:11 -04002824 dst_path = btrfs_alloc_path();
Tsutomu Itoh5df67082011-02-01 09:17:35 +00002825 if (!dst_path) {
2826 btrfs_free_path(path);
2827 return -ENOMEM;
2828 }
Chris Masone02119d2008-09-05 16:13:11 -04002829
Li Zefan33345d012011-04-20 10:31:50 +08002830 min_key.objectid = ino;
Chris Masone02119d2008-09-05 16:13:11 -04002831 min_key.type = BTRFS_INODE_ITEM_KEY;
2832 min_key.offset = 0;
2833
Li Zefan33345d012011-04-20 10:31:50 +08002834 max_key.objectid = ino;
Chris Mason12fcfd22009-03-24 10:24:20 -04002835
2836 /* today the code can only do partial logging of directories */
2837 if (!S_ISDIR(inode->i_mode))
2838 inode_only = LOG_INODE_ALL;
2839
Chris Masone02119d2008-09-05 16:13:11 -04002840 if (inode_only == LOG_INODE_EXISTS || S_ISDIR(inode->i_mode))
2841 max_key.type = BTRFS_XATTR_ITEM_KEY;
2842 else
2843 max_key.type = (u8)-1;
2844 max_key.offset = (u64)-1;
2845
Miao Xie16cdcec2011-04-22 18:12:22 +08002846 ret = btrfs_commit_inode_delayed_items(trans, inode);
2847 if (ret) {
2848 btrfs_free_path(path);
2849 btrfs_free_path(dst_path);
2850 return ret;
2851 }
2852
Chris Masone02119d2008-09-05 16:13:11 -04002853 mutex_lock(&BTRFS_I(inode)->log_mutex);
2854
2855 /*
2856 * a brute force approach to making sure we get the most uptodate
2857 * copies of everything.
2858 */
2859 if (S_ISDIR(inode->i_mode)) {
2860 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
2861
2862 if (inode_only == LOG_INODE_EXISTS)
2863 max_key_type = BTRFS_XATTR_ITEM_KEY;
Li Zefan33345d012011-04-20 10:31:50 +08002864 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
Chris Masone02119d2008-09-05 16:13:11 -04002865 } else {
2866 ret = btrfs_truncate_inode_items(trans, log, inode, 0, 0);
2867 }
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002868 if (ret) {
2869 err = ret;
2870 goto out_unlock;
2871 }
Chris Masone02119d2008-09-05 16:13:11 -04002872 path->keep_locks = 1;
2873
Chris Masond3977122009-01-05 21:25:51 -05002874 while (1) {
Chris Mason31ff1cd2008-09-11 16:17:57 -04002875 ins_nr = 0;
Chris Masone02119d2008-09-05 16:13:11 -04002876 ret = btrfs_search_forward(root, &min_key, &max_key,
2877 path, 0, trans->transid);
2878 if (ret != 0)
2879 break;
Chris Mason3a5f1d42008-09-11 15:53:37 -04002880again:
Chris Mason31ff1cd2008-09-11 16:17:57 -04002881 /* note, ins_nr might be > 0 here, cleanup outside the loop */
Li Zefan33345d012011-04-20 10:31:50 +08002882 if (min_key.objectid != ino)
Chris Masone02119d2008-09-05 16:13:11 -04002883 break;
2884 if (min_key.type > max_key.type)
2885 break;
Chris Mason31ff1cd2008-09-11 16:17:57 -04002886
Chris Masone02119d2008-09-05 16:13:11 -04002887 src = path->nodes[0];
Chris Mason31ff1cd2008-09-11 16:17:57 -04002888 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
2889 ins_nr++;
2890 goto next_slot;
2891 } else if (!ins_nr) {
2892 ins_start_slot = path->slots[0];
2893 ins_nr = 1;
2894 goto next_slot;
Chris Masone02119d2008-09-05 16:13:11 -04002895 }
2896
Chris Mason31ff1cd2008-09-11 16:17:57 -04002897 ret = copy_items(trans, log, dst_path, src, ins_start_slot,
2898 ins_nr, inode_only);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002899 if (ret) {
2900 err = ret;
2901 goto out_unlock;
2902 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04002903 ins_nr = 1;
2904 ins_start_slot = path->slots[0];
2905next_slot:
Chris Masone02119d2008-09-05 16:13:11 -04002906
Chris Mason3a5f1d42008-09-11 15:53:37 -04002907 nritems = btrfs_header_nritems(path->nodes[0]);
2908 path->slots[0]++;
2909 if (path->slots[0] < nritems) {
2910 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
2911 path->slots[0]);
2912 goto again;
2913 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04002914 if (ins_nr) {
2915 ret = copy_items(trans, log, dst_path, src,
2916 ins_start_slot,
2917 ins_nr, inode_only);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002918 if (ret) {
2919 err = ret;
2920 goto out_unlock;
2921 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04002922 ins_nr = 0;
2923 }
David Sterbab3b4aa72011-04-21 01:20:15 +02002924 btrfs_release_path(path);
Chris Mason3a5f1d42008-09-11 15:53:37 -04002925
Chris Masone02119d2008-09-05 16:13:11 -04002926 if (min_key.offset < (u64)-1)
2927 min_key.offset++;
2928 else if (min_key.type < (u8)-1)
2929 min_key.type++;
2930 else if (min_key.objectid < (u64)-1)
2931 min_key.objectid++;
2932 else
2933 break;
2934 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04002935 if (ins_nr) {
2936 ret = copy_items(trans, log, dst_path, src,
2937 ins_start_slot,
2938 ins_nr, inode_only);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002939 if (ret) {
2940 err = ret;
2941 goto out_unlock;
2942 }
Chris Mason31ff1cd2008-09-11 16:17:57 -04002943 ins_nr = 0;
2944 }
2945 WARN_ON(ins_nr);
Chris Mason9623f9a2008-09-11 17:42:42 -04002946 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
David Sterbab3b4aa72011-04-21 01:20:15 +02002947 btrfs_release_path(path);
2948 btrfs_release_path(dst_path);
Chris Masone02119d2008-09-05 16:13:11 -04002949 ret = log_directory_changes(trans, root, inode, path, dst_path);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002950 if (ret) {
2951 err = ret;
2952 goto out_unlock;
2953 }
Chris Masone02119d2008-09-05 16:13:11 -04002954 }
Chris Mason3a5f1d42008-09-11 15:53:37 -04002955 BTRFS_I(inode)->logged_trans = trans->transid;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002956out_unlock:
Chris Masone02119d2008-09-05 16:13:11 -04002957 mutex_unlock(&BTRFS_I(inode)->log_mutex);
2958
2959 btrfs_free_path(path);
2960 btrfs_free_path(dst_path);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04002961 return err;
Chris Masone02119d2008-09-05 16:13:11 -04002962}
2963
Chris Mason12fcfd22009-03-24 10:24:20 -04002964/*
2965 * follow the dentry parent pointers up the chain and see if any
2966 * of the directories in it require a full commit before they can
2967 * be logged. Returns zero if nothing special needs to be done or 1 if
2968 * a full commit is required.
2969 */
2970static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
2971 struct inode *inode,
2972 struct dentry *parent,
2973 struct super_block *sb,
2974 u64 last_committed)
Chris Masone02119d2008-09-05 16:13:11 -04002975{
Chris Mason12fcfd22009-03-24 10:24:20 -04002976 int ret = 0;
2977 struct btrfs_root *root;
Josef Bacik6a912212010-11-20 09:48:00 +00002978 struct dentry *old_parent = NULL;
Chris Masone02119d2008-09-05 16:13:11 -04002979
Chris Masonaf4176b2009-03-24 10:24:31 -04002980 /*
2981 * for regular files, if its inode is already on disk, we don't
2982 * have to worry about the parents at all. This is because
2983 * we can use the last_unlink_trans field to record renames
2984 * and other fun in this file.
2985 */
2986 if (S_ISREG(inode->i_mode) &&
2987 BTRFS_I(inode)->generation <= last_committed &&
2988 BTRFS_I(inode)->last_unlink_trans <= last_committed)
2989 goto out;
2990
Chris Mason12fcfd22009-03-24 10:24:20 -04002991 if (!S_ISDIR(inode->i_mode)) {
2992 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
2993 goto out;
2994 inode = parent->d_inode;
2995 }
2996
2997 while (1) {
2998 BTRFS_I(inode)->logged_trans = trans->transid;
2999 smp_mb();
3000
3001 if (BTRFS_I(inode)->last_unlink_trans > last_committed) {
3002 root = BTRFS_I(inode)->root;
3003
3004 /*
3005 * make sure any commits to the log are forced
3006 * to be full commits
3007 */
3008 root->fs_info->last_trans_log_full_commit =
3009 trans->transid;
3010 ret = 1;
3011 break;
3012 }
3013
3014 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
3015 break;
3016
Yan, Zheng76dda932009-09-21 16:00:26 -04003017 if (IS_ROOT(parent))
Chris Mason12fcfd22009-03-24 10:24:20 -04003018 break;
3019
Josef Bacik6a912212010-11-20 09:48:00 +00003020 parent = dget_parent(parent);
3021 dput(old_parent);
3022 old_parent = parent;
Chris Mason12fcfd22009-03-24 10:24:20 -04003023 inode = parent->d_inode;
3024
3025 }
Josef Bacik6a912212010-11-20 09:48:00 +00003026 dput(old_parent);
Chris Mason12fcfd22009-03-24 10:24:20 -04003027out:
Chris Masone02119d2008-09-05 16:13:11 -04003028 return ret;
3029}
3030
Chris Mason257c62e2009-10-13 13:21:08 -04003031static int inode_in_log(struct btrfs_trans_handle *trans,
3032 struct inode *inode)
3033{
3034 struct btrfs_root *root = BTRFS_I(inode)->root;
3035 int ret = 0;
3036
3037 mutex_lock(&root->log_mutex);
3038 if (BTRFS_I(inode)->logged_trans == trans->transid &&
3039 BTRFS_I(inode)->last_sub_trans <= root->last_log_commit)
3040 ret = 1;
3041 mutex_unlock(&root->log_mutex);
3042 return ret;
3043}
3044
3045
Chris Masone02119d2008-09-05 16:13:11 -04003046/*
3047 * helper function around btrfs_log_inode to make sure newly created
3048 * parent directories also end up in the log. A minimal inode and backref
3049 * only logging is done of any parent directories that are older than
3050 * the last committed transaction
3051 */
Chris Mason12fcfd22009-03-24 10:24:20 -04003052int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
3053 struct btrfs_root *root, struct inode *inode,
3054 struct dentry *parent, int exists_only)
Chris Masone02119d2008-09-05 16:13:11 -04003055{
Chris Mason12fcfd22009-03-24 10:24:20 -04003056 int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
Chris Masone02119d2008-09-05 16:13:11 -04003057 struct super_block *sb;
Josef Bacik6a912212010-11-20 09:48:00 +00003058 struct dentry *old_parent = NULL;
Chris Mason12fcfd22009-03-24 10:24:20 -04003059 int ret = 0;
3060 u64 last_committed = root->fs_info->last_trans_committed;
3061
3062 sb = inode->i_sb;
3063
Sage Weil3a5e1402009-04-02 16:49:40 -04003064 if (btrfs_test_opt(root, NOTREELOG)) {
3065 ret = 1;
3066 goto end_no_trans;
3067 }
3068
Chris Mason12fcfd22009-03-24 10:24:20 -04003069 if (root->fs_info->last_trans_log_full_commit >
3070 root->fs_info->last_trans_committed) {
3071 ret = 1;
3072 goto end_no_trans;
3073 }
3074
Yan, Zheng76dda932009-09-21 16:00:26 -04003075 if (root != BTRFS_I(inode)->root ||
3076 btrfs_root_refs(&root->root_item) == 0) {
3077 ret = 1;
3078 goto end_no_trans;
3079 }
3080
Chris Mason12fcfd22009-03-24 10:24:20 -04003081 ret = check_parent_dirs_for_sync(trans, inode, parent,
3082 sb, last_committed);
3083 if (ret)
3084 goto end_no_trans;
Chris Masone02119d2008-09-05 16:13:11 -04003085
Chris Mason257c62e2009-10-13 13:21:08 -04003086 if (inode_in_log(trans, inode)) {
3087 ret = BTRFS_NO_LOG_SYNC;
3088 goto end_no_trans;
3089 }
3090
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003091 ret = start_log_trans(trans, root);
3092 if (ret)
3093 goto end_trans;
Chris Mason12fcfd22009-03-24 10:24:20 -04003094
3095 ret = btrfs_log_inode(trans, root, inode, inode_only);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003096 if (ret)
3097 goto end_trans;
Chris Mason12fcfd22009-03-24 10:24:20 -04003098
Chris Masonaf4176b2009-03-24 10:24:31 -04003099 /*
3100 * for regular files, if its inode is already on disk, we don't
3101 * have to worry about the parents at all. This is because
3102 * we can use the last_unlink_trans field to record renames
3103 * and other fun in this file.
3104 */
3105 if (S_ISREG(inode->i_mode) &&
3106 BTRFS_I(inode)->generation <= last_committed &&
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003107 BTRFS_I(inode)->last_unlink_trans <= last_committed) {
3108 ret = 0;
3109 goto end_trans;
3110 }
Chris Masonaf4176b2009-03-24 10:24:31 -04003111
3112 inode_only = LOG_INODE_EXISTS;
Chris Masond3977122009-01-05 21:25:51 -05003113 while (1) {
Chris Mason12fcfd22009-03-24 10:24:20 -04003114 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
Chris Masone02119d2008-09-05 16:13:11 -04003115 break;
3116
Chris Mason12fcfd22009-03-24 10:24:20 -04003117 inode = parent->d_inode;
Yan, Zheng76dda932009-09-21 16:00:26 -04003118 if (root != BTRFS_I(inode)->root)
3119 break;
3120
Chris Mason12fcfd22009-03-24 10:24:20 -04003121 if (BTRFS_I(inode)->generation >
3122 root->fs_info->last_trans_committed) {
3123 ret = btrfs_log_inode(trans, root, inode, inode_only);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003124 if (ret)
3125 goto end_trans;
Chris Mason12fcfd22009-03-24 10:24:20 -04003126 }
Yan, Zheng76dda932009-09-21 16:00:26 -04003127 if (IS_ROOT(parent))
Chris Masone02119d2008-09-05 16:13:11 -04003128 break;
Chris Mason12fcfd22009-03-24 10:24:20 -04003129
Josef Bacik6a912212010-11-20 09:48:00 +00003130 parent = dget_parent(parent);
3131 dput(old_parent);
3132 old_parent = parent;
Chris Masone02119d2008-09-05 16:13:11 -04003133 }
Chris Mason12fcfd22009-03-24 10:24:20 -04003134 ret = 0;
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003135end_trans:
Josef Bacik6a912212010-11-20 09:48:00 +00003136 dput(old_parent);
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003137 if (ret < 0) {
3138 BUG_ON(ret != -ENOSPC);
3139 root->fs_info->last_trans_log_full_commit = trans->transid;
3140 ret = 1;
3141 }
Chris Mason12fcfd22009-03-24 10:24:20 -04003142 btrfs_end_log_trans(root);
3143end_no_trans:
3144 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04003145}
3146
3147/*
3148 * it is not safe to log dentry if the chunk root has added new
3149 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
3150 * If this returns 1, you must commit the transaction to safely get your
3151 * data on disk.
3152 */
3153int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
3154 struct btrfs_root *root, struct dentry *dentry)
3155{
Josef Bacik6a912212010-11-20 09:48:00 +00003156 struct dentry *parent = dget_parent(dentry);
3157 int ret;
3158
3159 ret = btrfs_log_inode_parent(trans, root, dentry->d_inode, parent, 0);
3160 dput(parent);
3161
3162 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04003163}
3164
3165/*
3166 * should be called during mount to recover any replay any log trees
3167 * from the FS
3168 */
3169int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
3170{
3171 int ret;
3172 struct btrfs_path *path;
3173 struct btrfs_trans_handle *trans;
3174 struct btrfs_key key;
3175 struct btrfs_key found_key;
3176 struct btrfs_key tmp_key;
3177 struct btrfs_root *log;
3178 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
3179 struct walk_control wc = {
3180 .process_func = process_one_buffer,
3181 .stage = 0,
3182 };
3183
Chris Masone02119d2008-09-05 16:13:11 -04003184 path = btrfs_alloc_path();
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00003185 if (!path)
3186 return -ENOMEM;
3187
3188 fs_info->log_root_recovering = 1;
Chris Masone02119d2008-09-05 16:13:11 -04003189
Yan, Zheng4a500fd2010-05-16 10:49:59 -04003190 trans = btrfs_start_transaction(fs_info->tree_root, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003191 if (IS_ERR(trans)) {
3192 ret = PTR_ERR(trans);
3193 goto error;
3194 }
Chris Masone02119d2008-09-05 16:13:11 -04003195
3196 wc.trans = trans;
3197 wc.pin = 1;
3198
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00003199 ret = walk_log_tree(trans, log_root_tree, &wc);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003200 if (ret) {
3201 btrfs_error(fs_info, ret, "Failed to pin buffers while "
3202 "recovering log root tree.");
3203 goto error;
3204 }
Chris Masone02119d2008-09-05 16:13:11 -04003205
3206again:
3207 key.objectid = BTRFS_TREE_LOG_OBJECTID;
3208 key.offset = (u64)-1;
3209 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
3210
Chris Masond3977122009-01-05 21:25:51 -05003211 while (1) {
Chris Masone02119d2008-09-05 16:13:11 -04003212 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003213
3214 if (ret < 0) {
3215 btrfs_error(fs_info, ret,
3216 "Couldn't find tree log root.");
3217 goto error;
3218 }
Chris Masone02119d2008-09-05 16:13:11 -04003219 if (ret > 0) {
3220 if (path->slots[0] == 0)
3221 break;
3222 path->slots[0]--;
3223 }
3224 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3225 path->slots[0]);
David Sterbab3b4aa72011-04-21 01:20:15 +02003226 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003227 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
3228 break;
3229
3230 log = btrfs_read_fs_root_no_radix(log_root_tree,
3231 &found_key);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003232 if (IS_ERR(log)) {
3233 ret = PTR_ERR(log);
3234 btrfs_error(fs_info, ret,
3235 "Couldn't read tree log root.");
3236 goto error;
3237 }
Chris Masone02119d2008-09-05 16:13:11 -04003238
3239 tmp_key.objectid = found_key.offset;
3240 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
3241 tmp_key.offset = (u64)-1;
3242
3243 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003244 if (IS_ERR(wc.replay_dest)) {
3245 ret = PTR_ERR(wc.replay_dest);
3246 btrfs_error(fs_info, ret, "Couldn't read target root "
3247 "for tree log recovery.");
3248 goto error;
3249 }
Chris Masone02119d2008-09-05 16:13:11 -04003250
Yan Zheng07d400a2009-01-06 11:42:00 -05003251 wc.replay_dest->log_root = log;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003252 btrfs_record_root_in_trans(trans, wc.replay_dest);
Chris Masone02119d2008-09-05 16:13:11 -04003253 ret = walk_log_tree(trans, log, &wc);
3254 BUG_ON(ret);
3255
3256 if (wc.stage == LOG_WALK_REPLAY_ALL) {
3257 ret = fixup_inode_link_counts(trans, wc.replay_dest,
3258 path);
3259 BUG_ON(ret);
3260 }
Chris Masone02119d2008-09-05 16:13:11 -04003261
3262 key.offset = found_key.offset - 1;
Yan Zheng07d400a2009-01-06 11:42:00 -05003263 wc.replay_dest->log_root = NULL;
Chris Masone02119d2008-09-05 16:13:11 -04003264 free_extent_buffer(log->node);
Chris Masonb263c2c2009-06-11 11:24:47 -04003265 free_extent_buffer(log->commit_root);
Chris Masone02119d2008-09-05 16:13:11 -04003266 kfree(log);
3267
3268 if (found_key.offset == 0)
3269 break;
3270 }
David Sterbab3b4aa72011-04-21 01:20:15 +02003271 btrfs_release_path(path);
Chris Masone02119d2008-09-05 16:13:11 -04003272
3273 /* step one is to pin it all, step two is to replay just inodes */
3274 if (wc.pin) {
3275 wc.pin = 0;
3276 wc.process_func = replay_one_buffer;
3277 wc.stage = LOG_WALK_REPLAY_INODES;
3278 goto again;
3279 }
3280 /* step three is to replay everything */
3281 if (wc.stage < LOG_WALK_REPLAY_ALL) {
3282 wc.stage++;
3283 goto again;
3284 }
3285
3286 btrfs_free_path(path);
3287
3288 free_extent_buffer(log_root_tree->node);
3289 log_root_tree->log_root = NULL;
3290 fs_info->log_root_recovering = 0;
3291
3292 /* step 4: commit the transaction, which also unpins the blocks */
3293 btrfs_commit_transaction(trans, fs_info->tree_root);
3294
3295 kfree(log_root_tree);
3296 return 0;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003297
3298error:
3299 btrfs_free_path(path);
3300 return ret;
Chris Masone02119d2008-09-05 16:13:11 -04003301}
Chris Mason12fcfd22009-03-24 10:24:20 -04003302
3303/*
3304 * there are some corner cases where we want to force a full
3305 * commit instead of allowing a directory to be logged.
3306 *
3307 * They revolve around files there were unlinked from the directory, and
3308 * this function updates the parent directory so that a full commit is
3309 * properly done if it is fsync'd later after the unlinks are done.
3310 */
3311void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
3312 struct inode *dir, struct inode *inode,
3313 int for_rename)
3314{
3315 /*
Chris Masonaf4176b2009-03-24 10:24:31 -04003316 * when we're logging a file, if it hasn't been renamed
3317 * or unlinked, and its inode is fully committed on disk,
3318 * we don't have to worry about walking up the directory chain
3319 * to log its parents.
3320 *
3321 * So, we use the last_unlink_trans field to put this transid
3322 * into the file. When the file is logged we check it and
3323 * don't log the parents if the file is fully on disk.
3324 */
3325 if (S_ISREG(inode->i_mode))
3326 BTRFS_I(inode)->last_unlink_trans = trans->transid;
3327
3328 /*
Chris Mason12fcfd22009-03-24 10:24:20 -04003329 * if this directory was already logged any new
3330 * names for this file/dir will get recorded
3331 */
3332 smp_mb();
3333 if (BTRFS_I(dir)->logged_trans == trans->transid)
3334 return;
3335
3336 /*
3337 * if the inode we're about to unlink was logged,
3338 * the log will be properly updated for any new names
3339 */
3340 if (BTRFS_I(inode)->logged_trans == trans->transid)
3341 return;
3342
3343 /*
3344 * when renaming files across directories, if the directory
3345 * there we're unlinking from gets fsync'd later on, there's
3346 * no way to find the destination directory later and fsync it
3347 * properly. So, we have to be conservative and force commits
3348 * so the new name gets discovered.
3349 */
3350 if (for_rename)
3351 goto record;
3352
3353 /* we can safely do the unlink without any special recording */
3354 return;
3355
3356record:
3357 BTRFS_I(dir)->last_unlink_trans = trans->transid;
3358}
3359
3360/*
3361 * Call this after adding a new name for a file and it will properly
3362 * update the log to reflect the new name.
3363 *
3364 * It will return zero if all goes well, and it will return 1 if a
3365 * full transaction commit is required.
3366 */
3367int btrfs_log_new_name(struct btrfs_trans_handle *trans,
3368 struct inode *inode, struct inode *old_dir,
3369 struct dentry *parent)
3370{
3371 struct btrfs_root * root = BTRFS_I(inode)->root;
3372
3373 /*
Chris Masonaf4176b2009-03-24 10:24:31 -04003374 * this will force the logging code to walk the dentry chain
3375 * up for the file
3376 */
3377 if (S_ISREG(inode->i_mode))
3378 BTRFS_I(inode)->last_unlink_trans = trans->transid;
3379
3380 /*
Chris Mason12fcfd22009-03-24 10:24:20 -04003381 * if this inode hasn't been logged and directory we're renaming it
3382 * from hasn't been logged, we don't need to log it
3383 */
3384 if (BTRFS_I(inode)->logged_trans <=
3385 root->fs_info->last_trans_committed &&
3386 (!old_dir || BTRFS_I(old_dir)->logged_trans <=
3387 root->fs_info->last_trans_committed))
3388 return 0;
3389
3390 return btrfs_log_inode_parent(trans, root, inode, parent, 1);
3391}
3392